Ebook Ipgdcon 2016 7Th Post Graduate Doctoral Conference For Information Technology PDF Full Chapter PDF

iPGDCon 2016 7th Post Graduate &
Doctoral Conference for Information

Technology - eBook PDF
Visit to download the full and correct content document:
https://ebooksecure.com/download/ipgdcon-2016-7th-post-graduate-doctoral-confere
nce-for-information-technology-ebook-pdf/
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
McGraw Hill Education (India) Private Limited

NEW DELHI
McGraw Hill Education Offices

New Delhi New York St Louis San Francisco Auckland Bogotá Caracas
Kuala Lumpur Lisbon London Madrid Mexico City Milan Montreal
San Juan Santiago Singapore Sydney Tokyo Toronto
��
Published by McGraw Hill Education (India) Private Limited,

P-24, Green Park Extension, New Delhi 110 016.
��
Copyright © 2016, by McGraw Hill Education (India) Private Limited.
No part of this publication may be reproduced or distributed in any form or by any means, electronic, mechanical, photocopying, recording, or
otherwise or stored in a database or retrieval system without the prior written permission of the publishers. The program listing (if any) may
be entered, stored and executed in a computer system, but they may not be reproduced for publication.
This edition can be exported from India only by the publishers,
McGraw Hill Education (India) Private Limited.
ISBN (13): 978-93-5260-179-0

ISBN (10): 93-5260-179-3
Managing Director: Kaushik Bellani

Director—Products (Higher Education and Professional): Vibha Mahajan
Manager—Product Development (B&E-BGR): Hemant K Jha
Senior Researcher—Product Development: Navneet Mehra
Head—Production (Higher Education and Professional): Satinder S Baveja
Assistant Manager—Production: Atul Gupta
General Manager—Production: Rajender P Ghansela
Manager—Production: Reji Kumar
Information contained in this work has been obtained by McGraw Hill Education (India), from sources believed to be reliable. However, neither
McGraw Hill Education (India) nor its authors guarantee the accuracy or completeness of any information published herein, and neither McGraw
Hill Education (India) nor its authors shall be responsible for any errors, omissions, or damages arising out of use of this information. This work
is published with the understanding that McGraw Hill Education (India) and its authors are supplying information but are not attempting to render
engineering or other professional services. If such services are required, the assistance of an appropriate professional should be sought.
Typeset at Tej Composers, WZ 391, Madipur, New Delhi 110 063, and printed at
Cover Design: Kapil Gupta

Cover Printer:
��
Dr. Ravindra C. Thool

(20 July 1963 to 27 Feb 2016)
Ex- Member - Board of Studies (Information Technology)
Savitribai Phule Pune University, Pune
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
� ��
� ��
� ��
� ��
��
� ��
��
� ��
� ��
��
� ��
��
� ��
��
� ��
��
� ��
� ��
��
� ��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� 4. ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
��
��
��
��
��
��
��
��
��
� ��
� ��
� ��
��
��
��
��
xxiv Contents
��

��
� ��
� ��
� � ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
��
��
��
� � ��
��
��
��
� � ��
��
� ��
� ��
� ��
� ��
� ��
� ��
��
��
��
��
��
��
��
��
��
��
� � ��
��
Contents xxv
��

��
��
� ��
� ��
� ��
� ��
� ��
� � ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
� ��
��
��
��
��
��
��
��
��
��
� ��
� ��
� ��
� ��
��
��
��
��
��
��
��
��
��
��
xxvi Contents
��

��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
� ��
� ��
� � ��
��
��
��
� ��
� ��
� � ��
��
��
��
��
��
� � ��w��
��
� � ��
Contents xxvii
��

��
��
� ��
� ��
� � ��
��
� ��
� ��
� � ��
��
��
��
��
��
� ��
� ��
� � ��
��
��
��
��
��
��
��
� ��
� ��
� � ��
��
��
��
��
��
��
��
��
��
��
��
��
��
��
� ��
� ��
� � ��
��
xxviii Contents
��

��
��
��
��
��
��
� ��
� ��
� � ��
��
Resource-aware Scheduling Through Improved
Speculative Execution
Anuj Sachan S.B. Deshmukh
PG Student, Dept of I.T. Assistant Professor,Dept of I.T.
Pune Institute of Computer Technology Pune Institute of Computer Technology
Pune, India Pune, India
anujsachan@gmail.com dshyam100@yahoo.com
Abstract - Scheduling of a task in MapReduce is one of the and aspects. In MapReduce, multiple tasks run parallel on
most important aspects. To improve job completion and cluster substantial, task duplication redundantly executes some tasks on
throughput MapReduce uses speculative execution. When a which other tasks are critically dependent [6]. Mapreduce
machine takes an unusually long time to complete a task, the so- presents a new strategy referred to as LATE (Longest
called straggler machine, will delay the job completion time and Approximate Time to End) in its Hadoop-0.21 implementation. It
degrade the cluster throughput significantly. Many research efforts
keeps watch on the progress rate of the tasks and estimates their
being undertaken to increase MapReduce performance like LATE
(Longest Approximate Time to End), Resource utilization, Data remaining time. Tasks with their progress rate below slow Task
Placement in Hadoop clusters, but some are inappropriate and Threshold are selected as backup candidates [7] and therefore the
some did not lever the situation like data skew, asynchronously one with the longest remaining time is given the best priority.
starting task and abrupt resources.These problems are handled via
smart speculative execution wherein slow task is backed-up on an II. MAPREDUCE MECHANISM
alternative machine with the hope that the back-up one can finish MapReduce is a framework composed of two primitives viz.
later. In this paper, a new set of resource-aware scheduling with the map and reduce. It consumes data as its input and sort it into
help of improved speculative execution strategy is proposed. This
small lumps of data. These slots are processed by map tasks in
proposed technique improvises speculative execution significantly
by decreasing the job completion time and improving the cluster
totally parallel and distributed fashions outcomes of map task are
throughput. This new resource-aware scheduling technique will furnished into reduce tasks. After execution uncompleted and
aim at improving resource utilization across machines while failed task are again executed till completion of tasks map reduce
observing completion time goals. frame work is eternally associated with key value pair in
mapping function. Mapper Function will do tasks of mapping key
Keywords: MapReduce, Straggler machine, Hadoop, value of input to intermediate key value pairs. Mapping begins
Speculative Execution with zero and end up to n number of pairs. In reduce functions
there are three ground functions. They are shuffle, sort and
I. INTRODUCTION
reduce
MapReduce may be a programming model that has been The architecture of MapReduce is shown in Fig. 1:
��
implementation (simply known as MapReduce) and therefore the
well-liked ASCII text file implementation Hadoop which may be
obtained, at the side of the HDFS classification system from the
Apache Foundation. One have to be compelled to write area unit
2 functions, known as Map and Reduce, whereas the system
manages the parallel execution, coordination of tasks that execute
Map or Reduce, and conjointly deals with the likelihood that one
in all these tasks can fail to execute.
Scheduling of job is an important feature to consider in
MapReduce. MapReduce [1] has rapidly gained fame in both
academia and industry. It been used in indexing [1],
bioinformatics [2], machine learning [3] etc.
Hadoop is a widely-used implementation of MapReduce
besides Hadoop; other data parallel systems including Dryad [4]
and Sector/Sphere [5] have been designed with different features Fig. 1:- MapReduce Architecture
2 7th Post Graduate & Doctoral Conference for Information Technology
III. RELATED WORK B. Resource-aware Schedulers

There is a lot of research undertaken by research community
in developing scheduling algorithms which are resource-aware, 1. PRISM: Fine-Grained Resource-Aware Scheduling for
data locality aware, energy and makespan aware and speculative MapReduce [9]
execution based which are described as table 1 as under
The MapReduce work is to schedule the tasks in different
Table 1: Various Schedulers levels. In a MapReduce technique, it is a collection of jobs and
can be scheduled concurrently on multiple machines, resulting in
S.NO VARIOUS reduction in job running time.
DESCRIPTION
. SCHEDULERS
Many organizations such as Google, Facebook, and
Data Locality aware schedulers agenda is to Yahoo, they refer MapReduce to process large amount of data.
Data Locality schedule the job whose input data is stored on the
1.
Aware Schedulers requesting node otherwise whose input data is
But they refer at task level to execute these data. At the task
nearest to the requesting node. level, performance and effectiveness become essential to daily
life. at the start the task level performs 2 parts one is mapper part
Speculative Execution copy the task of
Speculative
MapReduce is called as a backup task. This and different is reducer phase. In mapper part, it takes
2. Execution Based information blocks Hadoop distributed classification system and
backup task is then run on another machine for
Schedulers
finishing the computation faster. it maps, mix the information and hold on within the multiple
Resource Aware It is for improving the resource utilization based files. Then the second, reducer part can fetch information from
3.
Schedulers on workload of the clusters. mapper output and shuffle, kind the information in a very
Performance serialized manner [10]. And may be simply enforced into new
Scheduling schemes for performance-driven
4. Management Base
d Schedulers
running of the Map Reduce tasks. versions of Hadoop.
There are many task scheduling schemes which
Energy and are power, thermal and makespan aware so as to Disadvantages:
5. Makespan Aware reduce the total energy consumption on these i. At the task-level varying resources offer lower performance.
Schedulers High Performance Computing (HPC) ii. It is difficult for task-level scheduler to utilize the run-time
datacenters.
resources. So that it lowers job execution time while
In network-aware scheduling, scheduling is done executing.
Network aware with respect to current status of the network so as iii. The key issue is that when one task has completed in phase-
6.
Schedulers to achieve better efficiency in scheduling of
tasks.
level, subsequent phase of task is not scheduled
��
resource conflict then passed to next phase.
A. Speculative Execution
2. Dynamic Job Ordering and Slot Configurations for
�� MapReduce Workloads.
and fix slow-running tasks; instead, it tries to identify when a
task is running slower than expected and launches another, This paper focuses on the job ordering and map/reduce slot
equivalent, task as a backup. configuration issues for MapReduce production workloads that
The speculative task scheduling depends upon scheduler in run periodically in a data storehouse, where the average
Hadoop and it is based on multiple assumptions, one is that data execution time of map/reduce tasks for a MapReduce job can be
center can be homogeneous, and all tasks progress at same rate graphed from the past run, under the FIFO scheduling in a
(while some may be local, some remote, some more compute Hadoop cluster. Two concert metrics are taken into
intensive etc.), and all reduce function tasks are processed with consideration, i.e., makespan and total completion time[11].
same amount of data. It is useful in decreasing the response time Further proposed a new greedy job ordering algorithm
�� and a map/reduce slot configuration algorithm to reduce the
be schedule at right time. Other things which need to be makespan and total completion time altogether.
considered is while deciding and scheduling them would be the
competition of resources (network, CPU etc) created by Disadvantages:
speculative (nothing but duplicate) tasks and selecting node to i. Map/reduce slot configuration issues for online jobs under
run them. To minimize the job response time speculative FIFO scheduling.
execution is used. ii. It considers two other existing Hadoop schedulers such as
Fair scheduler and Capacity scheduler.
Resource-aware Scheduling Through Improved Speculative Execution 3
iii. �� task. This backup task is then run to different machine for
priority dependency, where the restraint job ordering implementation the task in time limit and so that the
approach will be considered. computation becomes more faster. In Hadoop, if a node is
available but is performing poorly, the condition is called a
straggler; MapReduce runs a speculative copy of its task (also
3. Dynamic MR: A Dynamic Slot Allocation Optimization
called a backup task) on ��
Framework for MapReduce Clusters.
computation faster. Hadoop monitors task progress using a
parameter called Progress score which is valued between 0 and
This paper projected a DynamicMR framework desiring to
1. In map, the Progress Score is the tiny proportion of input data
improve the performance of MapReduce workloads whereas
read. For a reduce task, the execution is divided into three
managing the fairness. It consists of 3 techniques, namely, DHSA
phases, each of which accounts for 1/3 of the score. Hadoop
(Dynamic Hadoop Slot Allocation), SEPB (Speculative
looks at the average Progress Score of each category of tasks
Execution Performance Balancing), Slot Prescheduling, all of
��
that targets on the slot utilization optimisation for MapReduce
execution. When a tasks Progress Score is less than the average
cluster from numerous views[12]. DHSA focuses on the slot
for its category minus 0.2, and the task has run for at least one
utilization maximization by allocating map (or reduce) slots to
minute, it is marked as a straggler.
map and reduce tasks dynamically. significantly, it doesn't have
The comparative study of Speculative Execution based
any assumption or need any prior-knowledge and might be used
scheduler is given at table no.2
for any types of MapReduce jobs (e.g., freelance or dependent
ones). 2 kinds of DHSA ar conferred, namely, PI-DHSA and PD- D. Comparison Parameters
DHSA, supported numerous levels of fairness. User will choose The main focus in this research work is being on smarter
either of them consequently. In distinction to DHSA, SEPB and speculative execution strategies; studied and compared existing
Slot Pre programming contemplate the potency optimisation for a speculative execution based scheduler on the basis of following
given slot utilization.ions of Hadoop. parameters. This has helped us in determining smarter
speculative execution strategy.
Disadvantages:
i. Map and reduce slots can exchange, and then small jobs 1. Data Skew
cannot be accomplish in time. Data skew refers to the non-uniform distribution in a dataset.
ii. No load-balancing feature in scheduling for map and reduce These Tasks do not always process the same amount of data and
stage. it may experience several types of data skew in MapReduce.
When the input data has some big records that cannot be
iii. If reduce tasks steal map slot, some local map task will
separated, the map tasks that process those records will process
become non-local task because of shortage of map slots. more statistics. Partitioning the intermediate data generated by
the map tasks unevenly will also lead to the partition skew
4. Improve MapReduce Performance through Data Placement in among the reduce tasks, characteristically when the distribution
Heterogeneous Hadoop Clusters of keys in the input data set is skewed.
The direct impact of data skew on parallel execution of
Two algorithms were implemented and integrated into complex database queries is the deprived load balancing leading
Hadoop HDFS. The first algorithm is initially used to distribute to high response time.
file into heterogeneous nodes in a cluster. After the algorithm is
initiated all file fragments of an input files are distributed to the 2. Resource Utilization
computing nodes[13]. To solve the data skew problem the second Increasing utilization of system resources, increase the
algorithm is used and it also restructures file fragments. There performance of system. Maximum Resource utilization is in
two cases in which file fragments must be reshuffled. First, the Dolly [14] because of proactive cloning. Due to proactive cloning
cluster expanded is widened by adding new nodes to the present it uses maximum memory utilization then SAMR (Self Self-
cluster. When, new data is affixed to an existing input file, file adaptive MapReduce Scheduling Algorithm) [15] and ESAMR
fragments distributed by the initial data placement algorithm can (an Enhanced Self-Adaptive Map Reduce scheduling algorithm)
be disrupted. [16]. Disk Utilization is very much high in Wrangler [17] as
compared to SAMR and ESAMR.
C. Speculative execution based schedulers
3. Reliability and Scalability:
Wrangler is more reliable then SAMR and ESMAR. LATE is
These Schedulers are based on speculative task scheduling. less reliable than others. ESMAR, SMAR are quite scalable.
Speculative Execution copies the task and then it backed up the
4. Speed farther than a node counter vector is from the separating

Highest performance speed is in Wrangler rather than MCP. hyperplane, higher are the chances of it belonging to the predict
For small tasks Dolly gives good speed performance than others. class. To obtain a prospect estimate of the prediction being
correct, the distance from the separating hyperplane is converted
5. Heterogentiy and Confidence Measures to a number in the series [0, 1]. These probabilities are obtained
�� -�� by appropriate logistic regression models to this distance.
is not robust to modeling errors. To ensure reliable predictions, All the algorithms are heterogeneous in nature. Heterogeneous
the concept of confidence gauge along with the prediction of means having cluster of resources with different configuration or
these linear models is introduced. Confidence measure to help different resources in a cluster.
decide if the prediction are accurate enough for preventing Only Hadoop-LATE has an average confidence measure.
stragglers by influencing the scheduling decision is needed. The
Table 2: Speculative Execution based Scheduler (Comparative Study)
Factors Techniques
Name Hadoop- Hadoop-LATE Dolly SAMR ESAMR Hadoop-MCP Wrangler
Default
Algorithm - Algorithm In this Cloning Algorithm uses It differentiates MCP takes the Using linear
speculatively of jobs is done. historical the historical cost modeling
executes the information to information on performance of technique
tasks that will adjust the stage each node and it predicts
cluster
finish farthest weigh�� of divides them stragglers
into the MapReduce into k clusters computing
prospect. tasks during using K-means resources into
task execution clustering account
time. algorithm
Method - - Proactive Reactive Reactive - Proactive
Data Skew - Yes - - - No -
Resource Utilization - - Highest Medium Medium - High
Reliability - Medium High High High Medium Highest
Scalability - - - Yes Yes - -
Speed - Low High Medium Medium - High
Heterogeneity - Yes - Yes Yes Yes Yes
Confidence - Medium - High High - High
Measure
IV. PROPOSED SYSTEM IMPLEMENTATION 2. A custom scheduling is being used to schedule task according
following criteria:
In this proposed system a new set of resource-aware a) Estimated time of completion
scheduling with the help of improved speculative execution b) Average execution time
strategy is proposed which improvises speculative execution c) List of all chunk of data in descending order by their size.
significantly by lessening the job completion time and d) Avoiding Straggler nodes.
also improves the cluster throughput by assigning task slots in 3. Sorting of data using TIMSORT algorithm [18]
the genus of assigning jobs. This new resource-aware scheduling As by default MapReduce uses quicksort, so instead of using
technique will aim at improving resource utilization across quicksort, Timsort algorithm will be used. By using Timsort it
machines while observing completion time goals. The proposed will give 10-20% improvement from default Hadoop during
system is divided into 4 modules as shown in Fig. 2. In an shuffle and sort phase.
implementation phase: 4. Finally comparison between default Hadoop and modified
1. Identifying the stragglers either by reading previous job logs Hadoop version with proposed configuration will be done.
or during run time using EWMA algorithm Straggler will be
identified either by reading previous job logs or during run time
using EWMA algorithm
EWMA (Exponentially Weighted Moving Average) scheme
which is expressed as fol��
��
��
tradeoff between stability and responsiveness. however effective
EWMA will be in predicting the method speed and therefore the
remaining time of a task, a kind job is run within the cluster.
Resource-aware Scheduling Through Improved Speculative Execution 5
will give 10-20% improvement from default Hadoop during

shuffle and sort phase.
VI. CONCLUSION
Straggler identification and it's handling is accomplished

using EWMA which takes care of Resource utilization, results
into improving the job completion time and improved
throughput. By default, Hadoop implements quicksort in shuffle
phase. Through literature, it is found that Timsort which is more
efficient than quicksort, if used, will definitely provide 10-20%
better performance than default Hadoop. EWMA predicts
Fig. 2:Proposed System
��
V. EXPERIMENTAL RESULTS straggler nodes to non-straggler. Thus improves the resource
utilization and improves performance of Hadoop scheduler.
In an implementation phase, clusters have formed of 4 virtual
machine consisting of 2 on each physical node with 20 GB RAM
REFERENCES
and 1.5 TB of disk space, as given in table 3. Multi-node Hadoop
v2.6.2 cluster has been deployed and performed WordCount [1] J. Dean, et al. �� Simplified Data Processingon Large
Application on Input data. The input dataset size used is 30 GB. �� pp. 137-150, 2004
Table 3: Hadoop Cluster Configuration [2] Qiu, et al. �� technologies for bioinformatics ��
Node Hardware Configuration ��ACM, 2009.
Master Intel-i5, 2.2GHz, 6GB RAM, 200 GB Disk Space [3] C. T. Chu, et al. �� reduce for machine learning on ��
Slave1 Intel-i5, 2.2GHz, 4GB RAM, 200 GB Disk Space ��
Slave2 Intel-i3,1.8GHz, 4GB RAM, 150 GB Disk Space [4] M. Isard, et al. �� distributed data parallel program from sequential b
Slave3 Intel-i3,1.8GHz, 2GB RAM, 150 GB Disk Space uilding �� SIGOPS, vol. 41, no. 3, pp. 59�72, March 2007.
After submitting job, the slower machine identified will be called [5] Y. Gu, et al. �� and Sphere: the design and implementation of a high
as straggler as per following: performance data �� PTRS Series A:Mathematical, Physical and Engi
neering Sciences, vol.367, no. 1897, pp. 2429�2445, 2009.
A. Identifying the stragglers either by reading previous job logs [6] I. Ahmad, et al. �� New Approach to Scheduling Parallel Programs Using
or during run time using EWMA algorithm Task �� in Proc of ICPP, IEEE Computer Society, pp. 47�
51,1994.
In this phase the remaining time to complete the task is estimated
by the sum of the remaining time left in each phase. Cp (i.e., the [7] M. Zaharia, A. Konwinski, A. D. Joseph, R. Katz, and IStoica, ��
mapreduce performance in heterogeneous �� In Proc. Of the
current phase), Estimate the remaining time of tasks 8th USENIX conferencen,2008.
[8] Qi Chen, Cheng Liu, and Zhen Xiao, �� MapReduce Performance
Using Smart Speculative Execution �� IEEE Transactions On Co
mputer, 2013.
[9] Qi Zhang, Mohamed Faten Zhani,Yuke Yang, Raouf �� PRISM: Fi
ne Grained Resource Aware Scheduling for �� IEEE Transactio
ns on Cloud Computing, 2015.
To estimate the backup time of a slow task, use the sum of [10] �� Prism: Allocation Of Resources In
est_timep [8] for each phase in this task as estimation. Phase Level Using Map Reduce In �� Journal of
Research In Science & Engineering, 2015.
[11] Shanjiang Tang, Bu Sung Lee, Bingsheng �� Dynam ic Job Ordering an
d Slot Configurations for MapReduc e �� IEEE Transactions On
Remaining time of the copy phase using by following Services Comput ing, 2015.
[12] Shanjiang Tang, Bu Sung Lee, Bingsheng He, �� ic MR:A Dynamic
Slot Allocation Optimization Framework for MapReduce Cl�� , IEE
E Transactions On Cloud Computing, 2014.
[13] J. Xie, S. Yin, X. Ruan, Z. Ding, Y. Tian, J. Majors, A Manzanares and X
. Qin, �� MapReduce Performance through Data Placement in He
In the equation above, the process speed copy is estimated by terogeneous Hadoop �� International Symposium on Parallel
weighted moving average (WMA). & Distrbuted Processing,USA, pp 19-23, April 2010.
[14] ��
B. Sorting of data using TIMSORT algorithm [18] ��
Symposium on Networked Systems Design and Implementation (NSDI
As by default MapReduce uses quicksort, so instead of using ��
quick sort, Timsort algorithm will be used. By using Timsort it
[15] ��- [17] Neeraja J. Yadwadkar, Ganesh Ananthanarayanan, and Randy Katz.
adaptive MapReduce Scheduling Algorithm In Heterogeneous Wrangler: Predictable and faster jobs using fewer resources. In
�� IEEE International Conference on Computer Proceedings of the ACM ��
and Information Technology (CIT 2010),2010. pages 26:1�26:14, New York, NY, USA, 2014. ACM.
[16] C. He ; Y. Lu, X. Sun �� - Adaptive [18] ��
�� Parallel and Distributed Systems sorting a �� https://databricks.com/bl og/2014/10/10/spa
(ICPADS), 2012 IEEE 18th International Conference,2012 rk-petabyte-sort.html.
Emotionizer: Sentence Based Emotion Recognition
System
Gaurav Dalvi Emmanuel M.
PG Student, Dept of Information Technology HOD,Dept of Information Technology
Pune, India. Pune, India
gauravdalvi63@gmail.com hodit@pict.edu
Abstract� �� not be for human-computer interaction to acknowledge human
inferred from statistics, it also requires the support of some reviews emotions. Emotions can be expressed in many ways of which
or opinions of customers to provide better feedback or decision maximum communication is through text. Emotion Analysis
making. As most of the population are able to express their area has become a booming topic where Machine are trained to
emotions in the form of text, research in textual analysis becomes a
analysis emotion have helped organizations to get a closer look
prime need. Initially, a comparative analysis of existing techniques
at their products, Customer Satisfaction, Psychological Study of
like Keyword-spotting approach, Statistical and machine learning
approach, Rule-Based approach and Hybrid Approach, Cause
a patient and much more. Among all this major of the crowd
Extraction Approach is done. This research work uses keyword- expressing way is through Social Networking Sites like
spotting approach for retrieving affect words and calculating the Facebook, Microblogs, etc. Social Networking sites have
overall sentence level emotions. Also to get more accuracy it uses become an influencer to the people. People are following these
some heuristic rules which work on intensifying words and get sites for decision making, as a mentor in different walks of life.
more insight of emotions in sentences. The use of Microblogs is growing fast in recent years. Since
users are able to update their content quickly, microblogging
Keywords� �� Emotion Analysis, Natural
service also acts as a hub of real-time news.
Language Processing.
The organization of the paper is as section II describes Key
INTRODUCTION Concepts of Emotion Recognition, section III Comparative
Analysis, IV Proposed System, section V Results and
��
Discussion and finally concluded by section VI Conclusion.
be in his speech, facial expression, body language. Emotion is
often tangled with mood, nature, personality, and motivation. EMOTION RECOGNITION KEY CONCEPTS
Definitions of emotions in the literature have been many and
varied. Kleinginna and Kleinginna suggested a formal definition A. ��
�� The famous Emotion Classification method is the Big Six,
and objective factors, mediated by neural and hormonal systems, ��
which can a) give rise to effective experiences such as emotional expressions (Ekman el al. 1969). The Big Six
awakening of feelings , happiness and unhappiness; b) generate emotions are happiness, sadness, fear, surprise, anger and
cognitive processes such as emotionally relevant perceptual disgust. They are thought to be basic in two ways: psychological
affect, appraisals, labeling processes; c) active widespread and biological. They do not contain other emotions as parts, and
physiological adjustments to the arousing conditions; and d) lead they are innate.
to behavior that is often but not always communicative, goal- Ekman found that an isolated preliterate tribe in New Gineau,
�� the Fore tended to associate facial expressions of the Big Six
The larger use of new computer-based media for emotions with the same kinds of situations with which people
communication and expression gave rise to an increasing need associate them in the west [2].
D. Emotion Recognition Resources.

B. Differentiating Factors of Sentiments and Emotions.
WordNet: It is a large lexical database of English. Nouns, verbs,
adjectives and adverbs are grouped into sets of cognitive
synonyms, each expressing a distinct concept. Words that are
found in close proximity to one another in the network are
semantically disambiguated [4].
SentiWordNet: It is a lexical resource for opinion mining. It
assigns to each synset of WordNet three sentiment scores:
positive, negative, neutral [5].
ConceptNet: It is a semantic network containing a lot of things
computers should know about the world, especially when
understanding text written by people. Without common sense, a
��
��
�� e computer
reader would not know what to think [6].
Cyc: It is a $25 million, 20 years project in Artificial
Intelligence. Its database, at the time of writing, contains
1,000,000 of these hand-entered facts and is maintained in an
entire room full of computers. Opencyc is an open source
Fig 1: Differentiating factors between affect, feelings, emotions, sentiments and
opinions.
version provided by Cycorp [7].
SentiSense: It is a concept based affective lexicon. It is intended
People make use of the terms emotions and sentiments to be used in sentiment analysis related tasks, especially in
�� polarity and intensity classification and emotion classification
that result from the combined influences of the biological, the [8].
�� However, they are dissimilar by the
duration in which they are experienced. Emotions are short COMPARATIVE STUDY
episodes of the brain, conscious activity and behavioral changes Approaches to textual emotion recognition and classification
while sentiments have been found to form and held for a longer could be categorized into Keyword-spotting Approach,
period. Sentiments are formed and pointed towards an object, Statistical and Machine Learning Approach, Rule-Based and
whereas emotions are not always pointed towards an object [1]. Hybrid Approach, Cause Extraction Approach.
A page from Martin Luther King's reading of Nietzsche says that An initial approach to emotion recognition is based on spotting
Sentiment is powerless emotion [3]. emotional keywords. Affective Reasoner by Elliott is one of the
initial approaches towards keyword spotting. Affective Reasoner
C. Emotion bearing terms maintains an unambiguous set of emotion words and uses
Emotion words: words which convey some form of emotion keyword spotting approach to identify emotions in the sentences
which can be HAPPY, SAD, ANGER, DISGUST, SURPRISE, [10].
FEAR. ��
Emoticons: It expresses � usually by means of punctuation supervised machine learning with the SNoW learning
marks (though it can include numbers and letters) � a person's architecture in order to classify emotion in text. Its main goal
feelings or mood. was to classify the emotional affinity sentences in the narrative
Intensifiers: Sometimes we feel happier than just happy. But it is ��
difficult to convey the level of emotions. People use intensifier appropriate expressive rendering of text-to-speech synthesis
�� [11].
UPPERCASE.
Emotionizer: Sentence Based Emotion Recognition System 9
Rule-Based and Hybrid Approaches uses heuristic rules for disgust, surprise). Let Vk be the initial set of emotional
identifying the negation and intensifiers coupled up with ��
keyword spotting approach. Anthony proposed a system for synsets are searched by using SimilarTo relationship in
Real-Time �� WordNet. The newly retrieved synsets are added to S k. As the
Time Text-to-�� synsets are obtained indirectly, some penalty p is being attached
communication by extracting emotion from the real-time typed to it.
��

text displaying on the screen appropriate facial expression
images in real time [12].
In recent years, a new approach of extracting cause of the where qki = number of emotional synsets for the given word i
emotion emerged. Weiyuan Li and Hua Xu proposed a system and the given emotion type k
The emotional weight wki , k ��E, i �� m are calculated

with finds emotion cause in the text. This technique performs a
deep level analysis of the text by identifying the cause from
where the emotion arises. Posts with complicated linguistic from each word for W.
patterns are challenging to deal with [13].
��
��
��
�
��
��
PROPOSED SYSTEM
�
Fig 2. shows the various components of the Proposed System.
��
��
��
System Overview
��
�� {h, s, a, f, d, sp} [2].
For Emoticon dictionary construction list of list of emoticons

used by most popular chat systems: Gtalk, Skype, MSN Messenger
and Yahoo! Messenger [2].
3) Tokenization
The text is broken down at the sentence level and further at
the word level.
4) Preprocessing
After tokenizing the collected data, preprocessing is
performed. It involves steps like: POS tagging, Stop Words
Fig 2: System Architecture. Removal and Stemming.
5) Lexicon generation
System Architecture consists of 8 modules: ConceptNet / WordNet is used for building the emotion
1) Data Collection words dictionary by using SimilarTo relationship. Also, vectors
The input to this module is Any Text of which emotion of retrieved words using above relationship are calculated and
analysis need to be performed. For testing purpose, Alm Affect stored.
dataset of fairytales is used. 6) Keyword-spotting
2) Word and Emoticon Lexicon dictionary construction. Affect keywords are searched at sentence levels and are
The technique used for Word Lexicon dictionary mapped with corresponding emotion weight vectors.
construction is based on a simple method. Initially, a small set of 7) Heuristic Rules
unambiguous words tagged with emotions is collected. Such The rules are applied to intensify the emotion words if there
�� are any adjectives, exclamation marks, etc.
Then, WordNet dictionary is searched for synsets of initial 8) Emotion Calculation
collection. Emotion synset is denoted as Sk �� Calculating the overall emotion of the sentence is done.
set of six emotional types (happiness, sadness, anger, fear, 9) Formatting of the result
The results are formatted in XML and JSON for efficient Dictionary. For the construction of Emotion Lexicon dictionary,
delivery to other applications. ConceptNet and some core annotated words belonging to
emotion states are used. After getting all the vectors associated
RESULTS AND DISCUSSION
with emotion words in sentence Emotion Calculation is
Emotionizer works on any text�� performed.
Annotated dataset is used for testing. Tokenization of text is
performed to convert text into sentence list. Each sentence list is Experiment 1: Data Collection
passed through the preprocessing step for stopwords removal, ��
stemming to filter the sentence and get core stem words. consists of stories submitted by B. Potter (19 stories), H.C.
Keyword spotting technique is used to label the emotion words Anderson (77 stories), Grimm (80 stories). It dataset is
in a sentence with emotion vectors in Emotion Lexicon
annotated across 6 different emotion as Angry (A), Disgusted part of speech. For Stopwords Removal, nltk provides a
(D), Fearful (F), Happy (H), Neutral (N), Sad (Sa), Surprise predefined set of stop words which is used. Finding the root
(Su) [15]. words is an important step as it helps to match the root words
in Keyword spotting module.
Few sentences of Ginger and Pickles story.
Once upon a time there was a village shop. Results of POS tagging:
The name over the window was "Ginger and Pickles." [('Once', 'RB'), ('upon', 'IN'), ('a', 'DT'), ('time', 'NN'), ('there',
It was a little small shop just the right size for Dolls -- 'EX'), ('was', 'VBD'), ('a', 'DT'), ('village', 'NN'), ('shop', 'NN'),
Lucinda and Jane Doll-cook always bought their groceries at ('.', '.')]
Ginger and Pickles. [('The', 'DT'), ('name', 'NN'), ('over', 'IN'), ('the', 'DT'),
('window', 'NN'), ('was', 'VBD'), ('``', '``'), ('Ginger', 'NNP'),
Experiment 2: Sentence Tokenization ('and', 'CC'), ('Pickles', 'NNP'), ('.', '.'), ("''", "''")]
For Sentence tokenization, various pickles are provided. As [('It', 'PRP'), ('was', 'VBD'), ('a', 'DT'), ('little', 'JJ'), ('small',
most of the text available are �� english�� 'JJ'), ('shop', 'NN'), ('just', 'RB'), ('the', 'DT'), ('right', 'JJ'),
used for tokenization. Result produce is a form of a list which ('size', 'NN'), ('for', 'IN'), ('Dolls', 'NNP'), ('--', ':'), ('Lucinda',
can be iterated to get each sentence. 'NNP'), ('and', 'CC'), ('Jane', 'NNP'), ('Doll-cook', 'NNP'),
('always', 'RB'), ('bought', 'VBD'), ('their', 'PRP$'), ('groceries',
Sentence Tokenization result: 'NNS'), ('at', 'IN'), ('Ginger', 'NNP'), ('and', 'CC'), ('Pickles',
['Once upon a time there was a village shop.The name over 'NNP'), ('.', '.')]
the window was "Ginger and Pickles."', 'It was a little small
shop just the right size for Dolls -- Lucinda and Jane Doll- Result after applying Stopwords Removal and Stemming.
cook always bought their groceries at Ginger and Pickles.'] [u'ont', 'upon', 'tim', 'vil', 'shop', '.']
['the', 'nam', 'window', '``', 'ging', 'pickl', '.', "''"]
Experiment 3: Word Tokenization ['it', 'littl', 'smal', 'shop', 'right', 'siz', 'dol', '--', 'lucind', 'jan',
For word tokenization focuses on two criteria such as white 'doll-cook', 'alway', 'bought', u'grocery', 'ging', 'pickl', '.']
spaces/periods, punctuation marks. It split off punctuation
marks other than periods to form words. Experiment 5: Keyword Spotting
Word Lexicon dictionary is used which consists of emotion
Word Tokenization result: words along with emotion vectors. Words from this dictionary
['Once', 'upon', 'a', 'time', 'there', 'was', 'a', 'village', are used to match words after word tokenization to get
�� respective vectors.
'Ginger', 'and', 'Pickles', '.', "''"]
['It', 'was', 'a', 'little', 'small', 'shop', 'just', 'the', 'right', 'size', Result of Keyword Spotting:
'for', 'Dolls', '--', 'Lucinda', 'and', 'Jane', 'Doll-cook', 'always', {'shop': ['0.0', '0.0', '0.0', '0.0', '0.16666666666666666',
'bought', 'their', 'groceries', 'at', 'Ginger', 'and', 'Pickles', '.'] '0.0']}{}
{'shop': ['0.0', '0.0', '0.0', '0.0', '0.16666666666666666', '0.0'],
Experiment 4: Preprocessing 'right': ['0.027777777777777776', '0.0', '0.0', '0.0', '0.0', '0.0']}
Preprocessing consists of three steps as Part Of Speech (POS)
Tagging, Stopwords Removal, Stemming. POS tagging is
done after sentence tokenization to get different associated
Emotionizer: Sentence Based Emotion Recognition System 11
Experiment 6: Emotion Calculation at Sentence Level It was a little small shop just the right size for Dolls --
Lucinda and Jane Doll-cook always bought their groceries at
Emotion vectors of all affect words (emotion bearing words)
Ginger and Pickles.
are summed up and a neutral threshold is used to identify [0.027777777777777776, 0.0, 0.0, 0.0,
neutral sentence. s0.16666666666666666, 0.0] n
Final Outcome after Emotion Calculation:
Once upon a time there was a village shop.
[0.0, 0.0, 0.0, 0.0, 0.16666666666666666, 0.0] n
The name over the window was "Ginger and Pickles."
[0.0, 0.0, 0.0, 0.0, 0.0, 0.0] n
rdnet.princeton.edu, 2015. [Online].Available: https://wordnet.p
rinceton.edu/wordnet/. [Accessed:24- Dec- 2015].
[6] Sentiwordnet.isti.cnr.it, "SentiWordNet", 2015. [Online].
CONCLUSION Available: http://sentiwordnet.isti.cnr.it/. [Accessed: 24- Dec-
2015].
Emotions are conveyed in the form of audio, video, textual Conceptnet5.media.mit.edu, "ConceptNet 5", 2015.
[7]
format. Out of which communication among people is mostly [Online]. Available:http://conceptnet5.media.mit.edu/.
[Accessed: 24- Dec- 2015].
done through a textual form which gives rise to research in "Artificial Intelligence", 2016.
[8] Psych.utoronto.ca,
this field. Different approaches to extract emotion are [Online]. Available:http://psych.utoronto.ca/users/reingold/cours
discussed. It is concluded that Rule-Based Approach can be es/ai/cyc.html. [Accessed: 16- Mar- 2016].
[9] Carrillo-de-Albornoz, L. Plaza and P. Gervás, "SentiSense: An
enriched by adding more unambiguous rules bagged up with easily scalable concept-based affective lexicon for sentiment
efficient lexicon word generator. Lexicon word generator can analysis", in 8th International Conference on Language
Resources and Evaluation (LREC 2012), p. 1, 2012.
be built by using some preliminary emotion words sorted by [10] Elliott, C. Davidson, "The affective reasoner: A process model of
experts and then applying the procedure to construct lexicon emotions in a multi-agent system", Ph.D, Northwestern
University Evanston, IL, USA, 1992.
synonyms. ��
[11]
Machine Learning for Text-Based E��
REFERENCES Human Language Technology Conf., Conf. Empirical Methods
in Natural Language Processing, pp. 579-586, 2005.
[1] M. D. Munezero, C. Montero, E. Sutinen and J. Pajunen, 'Are �� -to-Emotion Engine for
They Different? Affect, Feeling, Emotion, Sentiment, and [12]
Expressive Internet �� ts,
Opinion Detection in Text', IEEE Trans. Affective Comput., vol. Effects and Measurement of User Presence in Synthetic
5, no. 2, pp. 101-111, 2014. Environments, pp. 306-318, Ios Press, 2003.
[2] U. Krcadinac, P. Pasquier, J. Jovanovic and V. Devedzic,
[13] W. Li and H. Xu, 'Text-based emotion classification using
'Synesketch: An Open Source Library for Sentence-Based emotion cause extraction', Expert Systems with Applications,
Emotion Recognition', IEEE Trans. Affective Comput., vol. 4, no. vol. 41, no. 4, pp. 1742-1749, 2014.
3, pp. 312-325, 2013.
[14] Bird, Steven, Edward Loper and Ewan Klein (2009), Natural
[3] J. PRINZ, 'Which emotions are basic?', Emotion, Evolution, and Language Processing with Python��
Rationality, pp. 69-88, 2004. C. Alm, "Downloads for affect data", People.rc.rit.edu, 2016. [O
Quora.com, "What is the difference between sentiment and [15]
[4] nline]. Available: http://people.rc.rit.edu/~coagla/affectdata/.
emotion ? Compare - Quora", 2015. [Online]. [Accessed: 16- Mar- 2016].
Available:https://www.quora.com/What-is-the-difference-
between-sentiment-and-emotion-Compare.[Accessed: 23- Dec-
2015].
[5] P. University, "About WordNet WordNet About Wo rdNet", Wo
��
SVM ��
Himanshu Joshi A.M. Bagade
PG Student, Dept Of Information Technology Assistant Professor, Dept of Information Technology
Pune, India Pune, India
himanshuj25@gmail.com ambagade@pict.edu
Abstract� Wide spread use of Biometric supporting to deal with variation in illumination which can drastic change
consumer system is increasing day by day. Consumer devices like in face recognition, other challenges are face pose, expression
laptop, tablets and phones have potential to act as a biometric etc.
readers used for authentication. However, this rapidly emerging The existing system for biometric verification uses either
technology lead to external attacks especially like Replay attack. controlled environment setup or specialized hardware sensors.
Current approaches to counter replay attacks in this area are
These techniques are expensive and limit the widespread
inadequate or being performed over controlled environment
which are basically used to distinguish between live people and development of biometric system as consumer devices like
spoofing attempts. These systems also have specialized hardware laptop, tablets etc. are operated in highly uncontrolled
such as sensor with specific processor chip which is very high in environments with no expensive hardware attached to them,
cost and not in reach of normal user. so image based feature extraction (Hybrid approach) like
The paper proposes a challenge-response method using face color space with SVM method can be used because this
reflection with in-band digital watermarking to address replay technique is cheap, reliable and can work against many
attacks for face recognition on smart devices. Here challenge is of environment challenges like auto brightening, auto white
�� balancing, etc. This paper propose a consumer device face
and response of identification and validation is done through
recognition system which have in-band digital watermarking
��
under ideal condition, color reflection from the face may be [2,12] to address replay attacks and also produce good result
accurately classified in consumer devices. in uncontrolled environment.
Keywords�Biometric, Authentication, Replay attack, Digital This paper is presented as follows: Section II gives a
watermarking, In-band, Face Recognition general view on face recognition method. Section III provides
a detailed review of literature on different techniques used in
INTRODUCTION face recognition. Section IV gives introduction to the
proposed system used. Section V discusses different modules
Users hate passwords [1]! In modern world, the fear of and the algorithm used in those modules. Section VI discussed
losing personal information and valuables like money urge the intermediate results and the future outcome to be expected.
user to use passwords. But due to complexity rules and Section VII covers conclusion and future works.
different passwords for different purposes, leads user to forget
one or more of them. FACE RECOGNITION
Biometric system is used to identify a person based upon
his physical and behavioral characteristics. Some of the Face recognition method involves various techniques like
familiar methods of biometric are fingerprint recognition, iris pattern recognition, computer vision, computer graphics,
scanner, face recognition, handwriting verification and palm image processing etc.
geometry. Among all these technology, a rapidly growing In Fig.1 shown below, the first step is to capture image
technique is face recognition and it is one of most popular either online or offline then preprocessing is done, after that
application of image analysis. Face recognition plays a vital feature extraction is carried out by different image processing
role in accessing personal information, security and human algorithm used for features extraction and then extracted
machine interaction. The performance of many face feature are matched with database using classifier like SVM
recognition application in controlled environment reached to and finally final result is obtained.
satisfactory level, but still research is going on for face The different type of face detection and recognition methods
recognition technique which can deal the uncontrolled are shown in Fig 2.
environment. Some challenges of uncontrolled environment is
Face Recognition using Facial Color Reflection and SVM Classification 13
implement simple skin color algorithm for detecting skin

pixel. As human face color varies relative to the
illumination so pixel for skin region is detected using
normalized color histogram. Kjeldson and Kender [3]
propose HSV color space to separate skin region from
��
saturation (S), and value or brightness (V).
b. Based on Motion:
When use of video sequence is vailable then motion
information can be used to locate moving objects.Moving
shadows like face and body parts can be extracted by simply
Fig 1: Block diagram of face recognition method using classifier
frame differences gathered thresholding.
c. Based on edge detection:
Sakai et al. [4] proposed face detection based on
edges. Here, his work is based on analyzing line drawing
of faces from photographs, and aiming to locate facial
features
2. Active Shape Model:

Active Shape Model (ASMs) focuses on complex
non rigid features and aimed to automatically locate
landmarks points that define the shape of any
Fig 2: Different Face Detection Techniques statistically modeled object in an image. ASM is
classified into following.
ANALYSIS OF EXISTING WORKS
a. Deformable Templates:
A. KNOWLEDGE BASED METHOD: Deformable templates is introduced by Yuille et al. [5], he
In this method, human facial features parameters are incorporated global information of the eye to improve the
calculated like features of mouth, eyes, lips, face, nose etc. reliability of extraction process. Deformation is based on local
and their relationships like relative distance, intensity etc.
valley, edge, peak and brightness, other than face boundary,
B. Feature Based Method:
Feature Based Method are further classified as shown in facial features like eyes, nose, mouth and eyebrows extraction
Fig 2: is a great challenge here.
1. Low Level Analysis: 3. Feature Analysis:
These algorithms find structural features that exit
Based on visual feature in low levels like intensity, color, when pose, viewpoint and lighting conditions can
edge, motion etc. it is further classified:
a. Based on Skin color: change and use these to locate faces. These methods are
Human face has a vital feature know as skin color. also called face localization method.
Tracking a face using skin color as a feature has many
a. Viola Jones Method:
advantages like color processing is much faster than Paul Viola and Michael Jones [6] presented a fast and
other facial feature processing. Under favorable robust method for face detection and minimum
lightning condition, color is oriented invariant. The main computation time with high accuracy at that time when
problem of this color representation of face obtained by they presented their paper. This technique uses Haar like
camera is that it is influenced by factors like ambient feature that evaluated quickly through use of new image
light, object movement etc. Crowley and Coutaz [11] representation.
C. Template matching Approach: all location in the image. The result of calculating the
Template matching methods are used to correlate between distance from face space is a face map.
pattern in input image and stored standard patterns of a whole
face or face features to determine the presence of a face or
face features. Deformable and Predefined templates can be E. Hybrid approach:
used.
This approach uses both statistical pattern recognition
D. Image Based Approach:
Image based approach is further classified into techniques like SVMs and Feature Analysis techniques
1. Neural Network: like HSV color space.
As, face recognition is a two class pattern recognition HSV Color Space and Support Vector Machine (SVM):
problem, various neural network algorithm have been
proposed. Neural network for face detection is feasible Daniel F. Smith et al. [9] proposes a technique where he
uses local feature and whole face region to recognize a face. It
for training system to capture complex class face first convert the RGB captured images into HSV image and
patterns of conditional density. One drawback of neural then by using feature extraction it extract the color feature and
network is that network architecture has very much then compared with the SVM training data, if matched then
face is recognized if not then not matched. It uses frame
number of layers, number of nodes and learning rate to capture algorithm to improve the quality of color patterns
get superb performance. Feraud and Bernier [7] against illumination, ambient light etc. Hybrid method
proposed a detection method by using auto associative considered as best because it uses both local features character
as well as whole face region to detect or recognize a face.
neural networks, in this a five layer auto associated
network is able to perform nonlinear principal PROPOSED SYSTEM
component analysis
�� pupil
region, because bright reflection from the dark pupil region is
easy to recognize and can provide high level dissimilarity for
2. Statistical Method:
separating different color reflection, this is suggested by
Nishino and Nayer in 2004 [10]. From this technique it can be
a. Support Vector Machine (SVM): also stated that face also reflect color not much as eye pupil,
Osuna et al. [8] first introduces SVMs for face but can be used in studying face feature as it provide large
detection. SVMs uses new paradigm to train polynomial ��
pupil.
function, neural networks, or radial basis function (RBF) Using this approach a face recognition system is proposed
classifiers. It uses structural risk minimization to which uses information from the different reflected color from
minimize an upper bound on expected generalized error. face and then SVM is used to classify color and getting out a
Osuna et al. train SVM for large scale problem like face result. In this system different color are used because more the
variety in color greater the entropy. So, it is planned to use
detection. SVMs in wavelet domain can detect faces and colors displayed through monitor screen and the reflected
pedestrians color image is captured through camera. Then to avoid replay
. attack on system, a simple in-band digital watermarking
technique is used; this part of the system is called challenge
b. Principal Component Analysis (PCA): part. In response part, response of reflected color is calculated
Kirby and Sirivich in 1988 [13] introduced a and analyzed. If calculated response matches the order of
presented colors, one can conclude that the video was taken at
technique based on the concept of Eigen faces known as live instant and from particular device, rather than it is old
principal component analysis (PCA). PCA on a training video is replayed because it cannot contain the same color
set of face images is performed to generate the Eigen sequence. This system would robustly detect the spoofing
faces in face space. Faces area images are projected onto attack like printed photo when it is shown to camera as printed
photo have different color response and hence can be
the clustered and subspace. Similarly, for non- face identified by the system.
training images. To detect a face in scene, distance
between an image region and face space is computed for
Frame difference algorithm, it is proposed to convert RGB

image to HSV because HSV color space provide three spaces
namely Hue (H), Saturation (S) and brightness or Value (V)
and hence more information about the frame is generated.
ALGORITHM 2: FRAME DIFFERENCING ALGORITHM [9]
INPUT 1: Captured frame/ List of Displayed Color .
1. ��
OUTPUT: Face Recognized (YES/NO).
2. For all �� do,

�� //��
��
i
�� ;

ii Convert ��
Fig 3:Architecture Diagram
iii ��

� �
��

ALGORITHMS
�� //M��
��
IMAGE CAPTURING (Challane Part):
iv �� ; //��

User image is taken by the camera and then by using
� �
�� -band digital
watermarking, the frames are passed to Response ��
��
�
System.
��
Classify �� SVM(�� )
Image capturing algorithm may overcome the problem of
Automatic White Balancing (AWB) by interleaving each color v
frame with a period of no color to delay AWB. Auto 3. End for
resolution is handled by giving a settle down time before
capturing interleave different color frame 4. Return Face Recognized (YES/NO)
ALGORITHM 1: IMAGE CAPTURING ALGORITHM [9]:
INPUT: User image
OUTPUT: Captured frame and list of color sequence
1. Start camera and allow camera setting to settle. EXPERIMENTAL SETUP AND INTERMEDIATE
2. Display Black.
RESSULT
3. Capture and save four frames.
4. For four cycles do (four cycles is approx. 1 sec of video)
i Display random color and wait 100 msec. A. EXPERIMENTAL SETUP:
ii Capture and discard one frame. To experiment is performed on a laptop having
iii Capture and save three frames. Ubuntu 15.0, OpenCV 3.0 and pygame module, and the
iv Display Black and wait 100 msec.
v Capture and discard one frame. face recognition system uses the web camera to capture
vi Capture and save three frames. the images. Video resolution was set to 1024X780 pixels
5. End for.
6. Capture and save three frames. at 25 fps, and experiments were performed by flashing a
7. Return Captured frames; List of Displayed data series of color using laptop screen. The colors were
Frame and Color analysis (Response part): chosen randomly with each color interleaved by
In this module, the frames and colors are analyzed using displaying a blank screen.
��
classifier, match score is generated and then the person 1. Dataset:
authentication is done. Here a color is determined by choosing Here FRAUD1 dataset[9] is used. There are many dataset like
majority in the set of same colors frame after a set of three REPLAY-ATTACK dataset and CASIA-FASD dataset which
black frames and if no clear majority is there, then final color areaimedto detect live people versus spoofing attempts, but
from last frame of those set is chosen (reason may be camera they do not (and cannot) attempt to validate that the person is
still performing adjustment and it is more likely that last frame not. Hence, FRAUD1 is used in this experiment because it
color is more stabilized than other frames). The use of set of provide more reliable data set on the basis of person liveness
three or may be more frame is used to improve entropy and and also against spoofing attacks.
hence enhance the success rate to correctly classify a color. In The FRAUD1 dataset contains the following items:
1. Videos of objects using a computer tablet screen that

displays different colours to illuminate them. The objects
consist of a blank piece of white paper, a printed face on a
piece of paper, and ten live subjects.
2. Training data used to train the Support Vector
Machine. This training data was captured by reflecting
colours from a blank white sheet of paper, and capturing
the images using the HP pavilion web camera.
B. INTERMEDIATE RESULTS AND DISCUSSIONS:
Experiment 1: Capturing images:

In this, experiment phase HP pavilion web camera is used
and Ubuntu 15.0 working environment with python 3.0 is
used. Here, the user should face the camera and distance from
Fig 4:(a) Before Auto resolution adjustment by camera is done. (b) Before
camera and face must be about 30 centimetres or less, to get
Automatic White Balancing is done. (c) Black color is flashed and reflected
good illumination from the face.
response is captured. (d) Blue color is flashed and reflected response is
Experiment 2: Flashing different colours from screen captured. (e) Green color is flashed and reflected response is captured. (f) Red
OpenCv package and pygame module is used for color is flashed and reflected response is captured.
image processing and displaying different sequence of
color in screen respectively
Experiment 5: Implementing Frame difference algorithm:
Experiment 3: Integrating digital in-band watermarking In this part, the input is the captured frame and list of
Here, in-band digital watermarking is used to sequence of color displayed. Here, the captured frame
differentiate between live captured face frame and will be first converted to HSV(i.e. H for Hue, S for
replay frame from previous captured data. In this Saturation and V for Value) space from RGB (R for
watermarking is done by using discrete wavelength Red, G for Green and B for blue) color space. Then a
transformation technique. threshold of each color in its color space is calculated
Experiment 4: Implementing Image Capturing and hence for each color a separated matrix based on its
algorithm: Hue and Saturation is being formed
In this experiment, implementation of the image Experiment 6: Classifying Color using SVM
capturing algorithm given in Section V of this paper is In this part, a multi class SVM like LIBSVM
done Black frames are used in-between to differentiate classification of color is done. And resulted output like
between two different colors. And the output of this authenticate face will be done. The general structure of
algorithm i.e. captured frames and list of sequence of the training data is:
color displayed is passed to next module.
In the below figure, it can be easily seen that auto <data-line> ::= <colour><SPACE>
resolution and auto white balancing is done in figure 4- 1:<red-percent><SPACE>
2:<yellow-percent><SPACE>
(a), (b) and other responses like capturing color 3:<green-percent><SPACE>
reflection from camera for blue, black, green and red 4:<cyan-percent><SPACE>
screen illumination is shown. 5:<blue-percent><SPACE>
6:<magenta-percent><SPACE>
7:<black-percent><EOL>
The percentage is calculated by dividing the number of
pixels for each color channel, by the total of pixels calculated
in each of the six color channels. The detail structure of the
training data is taken from
the FRAUD1 dataset homepage http://www.itee.uq.edu.au./sa References

s/dataset.
After the training data is processed then testing of data is [1] A. Adams and M.A. Sasse, ��
done by passing different color matrixes to SVM as it is Communications of the ACM vol 42, no.12,pp 40- 46,1999.
Multi-Class SVM. Hence, the percentage of matched color is [2] ��
checked and if the matching score is greater than 70%, ��
Information Processing Systems, vol. 10, no. 2, pp. 256- 270, June
authentication is said to be yes otherwise authentication 2014.
fails. At last the final result, system gives a yes authentication [3] Kjeldsen. R. �� Proceeding
and allow user to proceed and if result is no then of the second International Conference on Automatic Face and Gesture
authentication fails error will be given to user. Recognition, pp 312-317, 1996,.
[4] �� and
CONCLUSION classification �� Proc.First USA- JAPAN
Computer conference, pp 2-7, 1972.
[5] A. L.Yuille, P. W.Hallinan, and D. S. Cohen, Feature extraction from
Biometric system are used in many places nowadays and faces using deformable templates,Int. J.Comput. Vision 8, pp.99�111,
the robustness and accuracy is the key concern. Face 1992.
Recognition is mainly used to recognize the face of user and [6] Paul Viola, Michael Jones, Robust Real-Time Face Detection,
International Journal of Computer Vision 57(2), 137�154, 2004.
effectively differentiate one from another and also reject the [7] R. Fe´raud, O.J. Bernier, J.-E. Villet, and M. Collobert, A Fast
fake or spoof faces like printed face etc. andAccuract Face Detector Based on Neural Networks��
The existing systems have disadvantages that many Trans.PatternAnalysis and Machine Intelligence, vol. 22, no. 1, pp. 42-
systems fails to provide good result due to auto resolution, 53,Jan. 2001.
[8] �� Training Support
automatic white balancing (AWB) of camera, hence they need VectorMachines: An Application to Face Detection�� Proc. IEEE
controlled environment or highly specialized sensor which are Conf.Computer Visionand Pattern Recognition, pp. 130-136, 1997.
very expensive. This proposed system is very robust and [9] �� Recognition
�� on Consumer Devices: Reflection on ��
on Information forensics and security, vol.1, no.1, April 2015.
�� [10] ��
�� conference on Computer Vision and Pattern Recognition (CVPR)
classification of colors using SVM. RGB image is converted vol1, pp. I-444-I-451, Jul 2004.
to HSV color space for more entropy and accurateness. The [11] Crowley, ��
one main disadvantage of this type system which still to be Robotics and Autonomous Systems, Vol 19, pp 347-358, 1997.
overcome is it needs dark room to capture image, yet the [12] �� morphing concept
for secure transmission of image data ��
indoor light only decrease its accuracy by little. Many Computer Science, vol. 6, no. 9, pp. 987-992, 2010.
research is still going on and many new technique can be seen �� Application of the Karhunen-
in future years as biometric is still developing area and attract Loe`veProcedure for the Characterization of Human Faces��
many researchers. Trans.PatternAnalysis and Machine Intelligence, vol. 12, no. 1, pp.
103-108,Jan. 1990.
Another random document with
no related content on Scribd:
CROQUETTES OF RICE. (ENTREMETS.)
Wipe very clean, in a dry cloth,

seven ounces of rice, put it into a
clean stewpan, and pour on it a quart
of new milk; let it swell gently by the
side of the fire, and stir it often that it
may not stick to the pan, nor burn;
when it is about half done, stir to it five
ounces of pounded sugar, and six Croquettes.
bitter almonds beaten extremely fine:
the thin rind of half a fresh lemon may
be added in the first instance. The rice must be simmered until it is
soft, and very thick and dry; it should then be spread on a dish, and
left until cold, when it is to be rolled into small balls, which must be
dipped into beaten egg, and then covered in every part with the
finest bread-crumbs. When all are ready, fry them a light brown in
fresh butter, and dry them well before the fire, upon a sieve reversed
and covered with a very soft cloth, or with a sheet of white blotting
paper. Pile them in a hot dish, and send them to table quickly.
Rice, 7 oz.; milk, 1 quart; rind of lemon: 3/4 hour. Sugar, 5 oz.
bitter almonds, 6: 40 to 60 minutes, or more. Fried, 5 to 7 minutes.
FINER CROQUETTES OF RICE. (ENTREMETS.)
Swell the rice in thin cream, or in new milk strongly flavoured with
vanilla or cocoa-nut; add the same ingredients as in the foregoing
receipt, and when the rice is cold, form it into balls, and with the
thumb of the right hand hollow them sufficiently to admit in the centre
a small portion of peach jam, or of apricot marmalade; close the rice
well over it; egg, crumb, and fry the croquettes as usual. As, from the
difference of quality, the same proportions of rice and milk will not
always produce the same effect, the cook must use her discretion in
adding, should it be needed, sufficient liquid to soften the rice
perfectly: but she must bear in mind that if not boiled extremely thick
and dry, it will be difficult to make it into croquettes.[136]
136. We must repeat here what we have elsewhere stated as the result of many
trials of it, that good rice will absorb and become tender with three times its
own bulk or measure of liquid. Thus, an exact half pint (or half pound) will
require a pint and a half, with an extremely gentle degree of heat, to convert
it into a thoroughly soft but firm mass; which would, perhaps, be rather too
dry for croquettes. A pint of milk to four ounces of rice, if well managed,
would answer better.
SAVOURY CROQUETTES OF RICE. (ENTRÉE.)
These are made with the same preparation as the casserole of

rice of Chapter XVIII., but it must be boiled very dry, and left to
become quite cold before it is used. A few spoonsful of rich white
sauce stirred into it when it is nearly tender, will improve it much.
Form and hollow the croquettes as directed in the last receipt; fill
them with a small portion of minced fowl, partridge, or pheasant in a
thick sauce, or with a stewed oyster or two cut in quarters; close the
rice perfectly over them; egg, and crumb the croquettes, fry and
serve them garnished with crisped parsley. French cooks mix
sometimes a little grated Parmesan cheese with the rice at the
moment it is taken from the fire, and roll the croquettes in more after
they are egged; they press this on and dip them again in egg, and
then into the crumbs. Raise the pan high above the fire when the
croquettes are lightly browned, that they may heat through; then
heighten the colour, and lift them out immediately.
RISSOLES. (ENTRÉE.)
This is the French name for small fried pastry of various forms,
filled with meat or fish previously cooked; they may be made with
brioche, or with light puff-paste, either of which must be rolled
extremely thin. Cut it with a small round cutter fluted or plain; put a
little rich mince, or good pounded meat, in the centre, and moisten
the edges, and press them securely together that they may not burst
open in the frying. The rissoles may be formed like small patties, by
laying a second round of paste over the meat, or like cannelons; they
may, likewise, be brushed with egg, and sprinkled with vermicelli,
broken small, or with fine crumbs. They are sometimes made in the
form of croquettes, the paste being gathered round the meat, which
must form a ball.[137]
137. If our space will permit, more minute directions for these, and other small
dishes of the kind, shall be given in the chapter of Foreign Cookery.
In frying them, adopt the same plan as for the croquettes, raising
the pan as soon as the paste is lightly coloured. Serve all these fried
dishes well drained, and on a napkin.
From 5 to 7 minutes, or less.
VERY SAVOURY ENGLISH RISSOLES. (ENTRÉE.)
Make the forcemeat No. 1, Chapter VIII., sufficiently firm with

unbeaten yolk of egg, to roll rather thin on a well-floured board; cut it
into very small rounds, put a little pounded chicken in the centre of
one half, moistening the edges with water, or white of egg, lay the
remaining rounds over these, close them securely, and fry them in
butter a fine light brown; drain and dry them well, and heap them in
the middle of a hot dish, upon a napkin folded flat: these rissoles
may be egged and crumbed before they are fried.
SMALL FRIED BREAD PATTIES, OR CROUSTADES OF VARIOUS
KINDS.
These may be either sweet or savoury, and many of them may be

so promptly prepared, that they offer a ready resource when an extra
dish is unexpectedly required. They should be carefully fried very
crisp, and of a fine equal gold colour, either in clarified marrow, for
which we give our own receipt, or in really good butter.
DRESDEN PATTIES, OR CROUSTADES.
(Very delicate.)
Pare the crust neatly from one or two French rolls, slice off the
ends, and divide the remainder into as many patties as the size of
the rolls will allow; hollow them in the centre, dip them into milk or
thin cream, and lay them on a drainer over a dish; pour a spoonful or
two more of milk over them at intervals, but not sufficient to cause
them to break; brush them with egg, rasp the crust of the rolls over
them, fry and drain them well, fill them with a good mince, or with
stewed mushrooms or oysters, and serve them very hot upon a
napkin; they may be filled for the second course with warm apricot
marmalade, cherry-jam, or other good preserve. This receipt came to
us direct from Dresden, and on testing it we found it answer
excellently, and inserted it in an earlier edition of the present work.
We name this simply because it has been appropriated, with many
other of our receipts, by a contemporary writer without a word of
acknowledgment.
TO PREPARE BEEF MARROW FOR FRYING CROUSTADES,
SAVOURY TOASTS, &C.
At a season when butter of pure flavour is often procured with

difficulty, beef-marrow, carefully clarified, is a valuable substitute for
it; and, as it is abundantly contained in the joints which are in
constant request for soup-making, it is of slight comparative cost in a
well managed kitchen. It is often thrown into the stock-pot by
careless or indolent cooks, instead of being rendered available for
the many purposes to which it is admirably adapted. Take it from the
bones as fresh as possible, put it into a white jar, and melt it with a
very gentle degree of heat at the mouth of the oven, or by the side of
the stove, taking all precaution to prevent its being smoked or
discoloured; strain it off, through a very fine sieve or muslin, into a
clean pan or pans, and set it aside for use. It will be entirely
flavourless if prepared with due care and attention; but, if dissolved
with too great a degree of heat, it will acquire the taste almost of
dripping. A small quantity of fine salt may be sprinkled into the pan
with it when it is used for frying.
SMALL CROUSTADES, OR BREAD PATTIES, DRESSED IN
MARROW.
(Author’s Receipt.)
Cut very evenly, from a firm stale loaf, slices nearly an inch and a
half thick, and with a plain or fluted paste-cutter of between two and
three inches wide press out the number of patties required,
loosening them gently from the tin, to prevent their breaking; then,
with a plain cutter, scarcely more than half the size, mark out the
space which is afterwards to be hollowed from it. Melt some clarified
beef-marrow in a small saucepan or frying-pan, and, when it begins
to boil, put in the patties, and fry them gently until they are equally
coloured of a pale golden brown. In lifting them from the pan, let the
marrow (or butter) drain well from them; take out the rounds which
have been marked on the tops, and scoop out part of the inside
crumb, but leave them thick enough to contain securely the gravy of
the preparation put into them. Fill them with any good patty-meat,
and serve them very hot on a napkin.
Obs.—These croustades are equally good if dipped into clarified
butter or marrow, and baked in a tolerably quick oven. It is well, in
either case, to place them on a warm sheet of double white blotting-
paper while they are being filled, as it will absorb the superfluous fat.
A rich mince, with a thick, well-adhering sauce, either of mutton and
mushrooms, or oysters, or with fine herbs and an eschalot or two; or
of venison, or hare, or partridges, may be appropriately used for
them.
SMALL CROUSTADES À LA BONNE MAMAN.
(The Grandmama’s Patties.)

Prepare the croustades as above, or use for them French rolls of
very even shape, cut in thick equal slices. If quite round, the crust
may be left on; mark each slice with a small cutter in the centre, dip
the croustades into butter or marrow, fry them lightly, or bake them
without permitting them to become very hard; empty, and then fill
them; dish them without a napkin, and pour some good brown gravy
round, but not over them.
Obs.—From being cooked without butter, these and the preceding
patties are adapted to a Jewish table.
CURRIED TOASTS WITH ANCHOVIES.
Fry lightly, in good butter, clarified marrow, or very pure olive oil,
some slices of bread, free from crust, of about half an inch thick, and
two inches and a half square; lift them on to a dish, and spread a not
very thick layer of Captain White’s currie-paste on the top; place
them in a gentle oven for three or four minutes, then lay two or three
fillets of anchovies on each, replace them in the oven for a couple of
minutes, and send them immediately to table. Their pungency may
be heightened by the addition of cayenne pepper, when a very hot
preparation is liked.
Obs.—We have spoken but slightly in our chapter of curries of
Captain White’s currie-paste, though for many years we have had it
used in preference to any other, and always found it excellent.
Latterly, however, it has been obtained with rather less facility than
when attention was first attracted to it. The last which we procured
directed, on the label of the jar, that orders for it should be sent per
post to 83, Copenhagen Street, Islington. It may, however, be
procured without doubt from any good purveyor of sauces and other
condiments. It is sold in jars of all sizes, the price of the smallest
being one-and-sixpence. We certainly think it much superior to any
of the others which we have tested, its flavour being peculiarly
agreeable.
TO FILLET ANCHOVIES.
Drain them well from the pickle, take off the heads and fins, lay
them separately on a plate, and scrape off the skin entirely; then
place them on a clean dish and with a sharp-edged knife raise the
flesh on either side of the back-bone, passing it from the tail to the
shoulders, and keeping it nearly flat as it is worked along. Divide
each side (or fillet) in two, and use them as directed for the
preceding toasts or other purposes. They make excellent simple
sandwiches with slices of bread and butter only; but very superior
ones when they are potted or made into anchovy butter.
SAVOURY TOASTS.
Cut some slices of bread free from crust, about half an inch thick
and two inches and a half square; butter the tops thickly, spread a
little mustard on them, and then cover them with a deep layer of
grated cheese and of ham seasoned rather highly with cayenne; fry
them in good butter, but do not turn them in the pan; lift them out,
and place them in a Dutch oven for three or [TN: missing word.]
minutes to dissolve the cheese: serve them very hot.
To 4 tablespoonsful of grated English cheese, an equal portion of
very finely minced, or grated ham; but of Parmesan, or Gruyère, 6
tablespoonsful. Seasoning of mustard and cayenne.
Obs.—These toasts, for which we give the original receipt
unaltered, may be served in the cheese-course of a dinner. Such
mere “relishes” as they are called, do not seem to us to demand
much of our space, or many of them which are very easy of
preparation might be inserted here: a good cook, however, will easily
supply them at slight expense. Truffles minced, seasoned, and
stewed tender in butter with an eschalot or two, may be served on
fried toasts or croûtons and will generally be liked.
TO CHOOSE MACCARONI AND OTHER ITALIAN PASTES.
The Naples maccaroni, of which the pipes are large, and

somewhat thin, should be selected for the table in preference to the
Genoa, which is less in size, but more substantial, and better suited
to the formation of the various fanciful timbales[138] for which it is
usually chosen. We have inserted here no receipts for these,
because unless very skilfully prepared they are sure to fail, and they
are not in much request in this country, unless it be at the tables of
the aristocracy, for which they are prepared by efficient cooks. Of the
ribbon maccaroni (or lazanges) we have given particulars in the
pages which follow. The macaroncini, though not much larger than a
straw, requires much boiling for its size, to render it soft. The celery-
maccaroni is made very large and of an ornamental form, but in
short lengths. It is used by “professed” cooks as a sort of crust or
case for quenelle-forcemeat, or other expensive preparations of the
same nature. The ring or cut maccaroni is another form given to the
Italian paste: it may be had at almost any good foreign warehouse.
138. For an explanation of the term timbale, the reader is referred to the glossary
at the commencement of this volume.
All these pastes should be of a yellowish tint (by no means white

as one sees them when they are of inferior quality); they should also
be quite fresh, as they contract a most unpleasant flavour from being
too long stored. The Naples vermicelli, which is much larger than any
other, may be dressed like maccaroni: by many persons it is also
preferred to the smaller varieties for serving in soup.
TO BOIL MACCARONI.
We have always found the continental mode of dressing

maccaroni the best. English cooks sometimes soak it in milk and
water for an hour or more, before it is boiled, that the pipes may be
swollen to the utmost, but this is apt to render it pulpy, though its
appearance may be improved by it. Drop it lightly, and by degrees,
into a large pan of fast-boiling water, into which a little salt, and a bit
of butter the size of a walnut, have previously been thrown, and of
which the boiling should not be stopped by the addition of the
maccaroni. In about three-quarters of an hour the Naples maccaroni
will be sufficiently tender: every kind should always be perfectly
cooked, for otherwise it will prove very indigestible, but the pipes of
that commonly served should remain entire. Pour it into a large
cullender, and drain the water well from it. It should be very softly
boiled after the first minute or two.
Time of boiling:—Naples maccaroni, about 3/4 hour; Genoa,
nearly or quite 1 hour; macaroncini, 20 to 25 minutes; cut maccaroni,
10 minutes; Naples vermicelli (in water), about 20 minutes; longer in
soup, or milk.
RIBBON MACCARONI.[139]
139. The best ribbon-maccaroni which we have ever had, was from Mr. Cobbett’s,
18, Pall Mall. It is rather higher in price than the pipe maccaroni, but swells
so much in the boiling that a large quantity of it is not required for a dish. We
ought to add that Mr. Cobbett’s is not a professedly cheap house, but that all
he supplies is of excellent quality.
This kind of maccaroni, though more delicate in flavour and much

more quickly boiled than the pipe maccaroni, is far less frequently
seen at English tables; yet it is extremely good in many simple forms
and very wholesome, therefore well suited to invalids and children as
well as to persons in health. Drop it gradually into plenty of boiling
water, and turn it over occasionally that it may be equally done. Drain
it thoroughly when it is perfectly tender, and serve it quickly either
quite plain, to be eaten instead of vegetables or rice; or with a
compote of fruit; or with sugar and cinnamon, or lemon juice; or
prepared in any of the modes indicated for the Naples maccaroni.
To be boiled 15 to 18 minutes.
DRESSED MACCARONI.
After careful and repeated trial of different modes of dressing

various kinds of maccaroni, we find that in preparing them with
Parmesan cheese, unmixed with any of a more mellow nature, there
is always a chance of failure, from its tendency to gather into lumps;
we would therefore recommend the inexperienced reader to
substitute for it in part, at least, any finely flavoured English cheese;
and the better to ensure its blending smoothly with the other
ingredients (when neither white, nor any other thickened sauce is
used with it), to dissolve the butter, and to stir to it a small
teaspoonful of flour, before any liquid is added, then carefully to mix
with it the cream or gravy, as directed for Sauce Tournée, Chapter V.,
and to give this a boil before the maccaroni and cheese are added: if
gently tossed as these become hot, the whole will be smooth, and
the cheese will adhere properly to the paste. Four ounces of pipe
maccaroni is sufficient for a small dish, but from six to eight should
be prepared for a family party where it is liked. The common English
mode of dressing it is with grated cheese, butter, and cream, or milk.
French cooks substitute generally a spoonful or two of very strong
rich jellied gravy for the cream; and the Italians, amongst their many
other modes of serving it, toss it in rich brown gravy, with sufficient
grated cheese to flavour the whole strongly; they send it to table also
simply laid into a good Espagnole or brown gravy (that drawn from
the stufato,[140] for example), accompanied by a plate of grated
cheese. Another, and an easy mode of dressing it is to boil and drain
it well, and to put it into a deep dish, strewing grated cheese on
every layer, and adding bits of fresh butter to it. The top, in this case,
should be covered with a layer of fine bread-crumbs, mixed with
grated cheese; these should be moistened plentifully with clarified
butter, and colour given to them in the oven, or before the fire; the
crumbs may be omitted, and a layer of cheese substituted for them.
An excellent preparation of maccaroni may be made with any well-
flavoured, dry white cheese, which can be grated easily, at much
less cost than with the Parmesan, which is expensive, and in the
country not always procurable: and we think that the brown gravy
and a seasoning of cayenne are great improvements to it.
140. See Chapter of Foreign Cookery.
Maccaroni, 6 oz.; butter, 3 oz.; Parmesan (or other) cheese, 6 oz.;
cream, 4 tablespoonsful.
Obs.—Less of butter and cheese can be used by the strict
economist.
MACCARONI À LA REINE.
This is a very excellent and delicate mode of dressing maccaroni.

Boil eight ounces in the usual way, and by the time it is sufficiently
tender, dissolve gently ten ounces of any rich, well flavoured white
cheese in full three-quarters of a pint of good cream; add a little salt,
a rather full seasoning of cayenne, from half to a whole saltspoonful
of pounded mace, and a couple of ounces of sweet fresh butter. The
cheese should, in the first instance, be sliced very thin, and taken
quite free of the hard part adjoining the rind; it should be stirred in
the cream without intermission until it is entirely dissolved, and the
whole is perfectly smooth: the maccaroni, previously well drained,
may then be tossed gently in it, or after it is dished, the cheese may
be poured equally over the maccaroni. The whole, in either case,
may be thickly covered before it is sent to table, with fine crumbs of
bread fried of a pale gold colour, and dried perfectly, either before
the fire or in an oven, when such an addition is considered an
improvement. As a matter of precaution, it is better to boil the cream
before the cheese is melted in it; rich white sauce, or béchamel,
made not very thick, with an additional ounce or two of butter, may
be used to vary and enrich this preparation. If Parmesan cheese be
used for it, it must of course be grated; but, as we have said before,
it will not easily blend with the other ingredients so as to be smooth.
A portion of Stilton, free from the blue mould, would have a good
effect in the present receipt. Half the quantity may be served.
Maccaroni, 1/2 lb.; cheese, 10 oz.; good cream, 3/4 pint (or rich
white sauce); butter, 2 oz. (or more); little salt, fine cayenne, and
mace.
SEMOULINA AND POLENTA À L’ITALIENNE. (GOOD.)
(To serve instead of Maccaroni.)

Throw into a quart of milk, when it is fast boiling, half
a teaspoonful of salt, and then shake lightly into it five
ounces of the best semoulina; stir the milk as this is
added, and continue to do so from eight to ten minutes,
letting the mixture boil gently during the time. It should
be very thick, and great care must be taken to prevent
its sticking to the saucepan, which should be placed
over a clear fire on a bar or trivet, but not upon the
coals. Pour the semoulina, when it is done, into a
basin, or a plain mould which it will not fill by an inch or
two, and let it remain some hours in a cool place, that it
may become perfectly cold; it will then turn out quite
solid, and like a pudding in appearance. Cut it with a
large, sharp carving-knife, or a bit of thin wire, into half-
inch slices; wash the basin into which it was poured at
first, and butter it well; grate from six to eight ounces of Maize.
good cheese (Parmesan, or any other), and mix with it
a half-teaspoonful of cayenne, and twice as much
pounded mace; clarify from two to three ounces of fresh butter, and
put a small quantity into the basin, strew in a little of the cheese, and
then lay in the first slice of the semoulina, on this put a thick layer of
the cheese, moisten it with some drops of butter, and place the
second slice upon it; then more cheese and butter, and continue thus
until all the semoulina is replaced in the basin; put plenty of cheese
upon the top, add the remainder of the clarified butter, and bake the
mixture for about half an hour in a gentle oven. It should be of a fine
golden colour when served. Turn it carefully into a dish, and send it
instantly to table. A little rich brown gravy poured round might, to
some tastes, improve it, but it is excellent without, and may be
substituted for maccaroni, which it much resembles in flavour. It may
be enriched by adding butter to the milk, or by mixing with it a portion

Ebook Ipgdcon 2016 7Th Post Graduate Doctoral Conference For Information Technology PDF Full Chapter PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Ebook Ipgdcon 2016 7Th Post Graduate Doctoral Conference For Information Technology PDF Full Chapter PDF

Загружено:

Авторское право:

Доступные форматы

iPGDCon 2016 7th Post Graduate &

Doctoral Conference for Information

McGraw Hill Education (India) Private Limited

McGraw Hill Education Ofﬁces

Published by McGraw Hill Education (India) Private Limited,

ISBN (13): 978-93-5260-179-0

Managing Director: Kaushik Bellani

Cover Design: Kapil Gupta

Dr. Ravindra C. Thool

����� ������������������������������������������������������������������������������� ���

������ ��������������������������������������������� ���

������ ����������������������������������������������������������������������� ���

������ �������������������������������������������������������� ���

III. RELATED WORK B. Resource-aware Schedulers

4. Speed farther than a node counter vector is from the separating

will give 10-20% improvement from default Hadoop during

Straggler identification and it's handling is accomplished

D. Emotion Recognition Resources.

��� � ��� � �� ����� �� �� �� �� ���� � � ���� � � ��� �� � ���

The emotional weight wki , k ��E, i �� ������ m are calculated

���������������� {h, s, a, f, d, sp} [2].

For Emoticon dictionary construction list of list of emoticons

implement simple skin color algorithm for detecting skin

2. Active Shape Model:

Frame difference algorithm, it is proposed to convert RGB

2. For all �� � ������� ����� do,

���� ��� ���� ���;

iii �� ��� � ����� ���� ���� ������� �� �

�� �� � � ��� ��� ����� � ��

iv �� ��� � � ���� ���� �� ��� ��; //�� ��� �� �

1. Videos of objects using a computer tablet screen that

B. INTERMEDIATE RESULTS AND DISCUSSIONS:

Experiment 1: Capturing images:

the FRAUD1 dataset homepage http://www.itee.uq.edu.au./sa References

Wipe very clean, in a dry cloth,

These are made with the same preparation as the casserole of

Make the forcemeat No. 1, Chapter VIII., sufficiently firm with

These may be either sweet or savoury, and many of them may be

At a season when butter of pure flavour is often procured with

(The Grandmama’s Patties.)

The Naples maccaroni, of which the pipes are large, and

All these pastes should be of a yellowish tint (by no means white

We have always found the continental mode of dressing

This kind of maccaroni, though more delicate in flavour and much

After careful and repeated trial of different modes of dressing

This is a very excellent and delicate mode of dressing maccaroni.

(To serve instead of Maccaroni.)

Вам также может понравиться

��

��

��

��

��

The emotional weight wki , k ��E, i �� m are calculated

�� {h, s, a, f, d, sp} [2].

2. For all �� do,

�� ;

iii ��

��

iv �� ; //��