Вы находитесь на странице: 1из 48

Vexo

we make it
happen
Hello!
We are Team Vexo
We are here to give a
presentation on our project
Vexo – Handwriting
Recognition using Machine
Learning
Handwriting Recognition

Handwriting recognition (or HWR) is the ability of a


computer to receive and interpret intelligible
handwritten input from sources such as paper
documents, photographs, touch-screens and other
devices. The image of the written text may be sensed
"off line" from a piece of paper by optical scanning
(optical character recognition) or intelligent word
recognition. Alternatively, the movements of the pen
tip may be sensed "on line", for example by a pen-
based computer screen surface, a generally easier task
as there are more clues available.
Handwriting Recognition Softwares
Let’s start with

iSkysoft
iSkysoft is a perfect OCR tool for
PDF files. It can automatically
recognize scanned PDF and
make it editable with built-in
editing tools. And it provides
several OCR languages. Besides,
you can easily edit your PDF
texts, images, links and other
Key Features
➜Advanced OCR function ➜Easy provision of
allows you to convert and markups and adding
edit scanned PDFs easily. annotations, wherever
necessary.
➜Editing PDF texts, images,
and links is as easier as ➜You can easily create
making changes in Word. PDFs from a wide range of
document formats.
➜Add signature, password,
watermarks, signs, free- ➜You can also convert the
hand shapes in PDFs with PDF file to other formats
ease. like Excel, MS Word and
more.
Pros Cons

➜ Available for ➜ No dedicated


Mac and mobile app
Windows ➜ Only paid
➜ Great PDF version
creation and available
editing feature
➜ Significantly
less expensive
TopOCR
TopOCR is designed to be simple and
user-friendly for scanning books and
magazines with document cameras
and scanners. It combines a full
featured Image Editor and Word
Processor with advanced multi-core
image processing and three different
OCR engines. For document cameras,
Pros Cons

➜ 60+ ➜ Only online


recognition version
language available
➜ Support major ➜ Free version is
image formats not available
now
➜ Multi column
document
analysis
MyScript
MyScript is the market leader in
accurate, high-performance
handwriting recognition and digital
ink management software technology.
MyScript technology combines digital
ink management with easy searching
of handwritten text, as well as the
accurate recognition of complex
Key Features

➜Use ICR technology


➜ Interactive note
taking application
➜Supporting over 99
languages,
mathematical
equations, geometric
Pros Cons

➜Available on ➜ MyScript
both mobile cannot use
bitmapped
and desktop
input obtained
operating as image data
system from scanners
or cameras
➜Interactive
note taking ➜ Expensive
Google
Handwriting
Input
Google Handwriting Input is
handwriting recognition
software developed by
Google, which works in
touch input devices. It is
basically designed for
android smartphones.
Google Handwriting
Input is an ICR
Key Features

➜A useful complement to
touchscreen typing or
voice input
➜A fun way to enter
emojis by drawing
➜Useful for languages that
can be challenging to type
on a standard keyboard
➜Works across Android
Pros Cons

➜Free to use ➜There’s a


➜A useful slight delay
complement while the
to app
touchscreen translates
typing or scratch into
voice input actual typed
text.
➜A fun way to
Design Thinking
Framework
persona
Pharmace
utist Data
Entry
journal
ists
Police
officers
Noveli
st
Identifying Users and User Behavior

Online survey has been through


Google form to identify users
and user behavior

23 people participated in survey


Users

6% 1%4%
6%
22%
7%

22%
17%

15%

Pharmaceutist Novelist Police officers


Data entry workers Journalists Students
Teachers Bank Employees others
16

14

12

10

0
User persona non user persona

user persona buyer persona anti persona


Customer
Experience

➜Made tasks faster


and easier
➜Not available in
regional
language
Problem statement

Most of the available


applications are expensive

Free software have poor


performance

Most of the available software


do not support regional
Now let’s see

Vexo
Handwriting recognisation
software
BIG
CONCEPT
Vexo is incoparated
with machine learning
and neural network
algorithm
The main tasks of
the application is
to provide a
solution for
handwriting
recognition based
on touch input,
handwriting
recognition from
To be implemented
using perceptron
architecture, learning
parameters and
optimization algorithms
Algorithms
used is neural Implementaion
network environment
model and
deep learning Torch
are
To be
Implemented
Torch comes with a
large ecosystem of
community- driven
packages in machine
learning, computer
vision, signal
processing, parallel
processing, image,
video, audio and
We have been doing our work in GitHub
which helped us to seek help from other
professionals in Neural Network and
Machine Learning
GitHub is a development platform
inspired by the way you work.
From open source to business, we
can host and review code, manage
projects, and build software alongside
millions of other developers.

GitHub brings teams together to work


through problems, move ideas forward,
and learn from each other along the
way.
Prototype

Currently Vexo is under


progress. Once the prototype is
ready Vexo has to be taught
and tested with handwritten
digits in “THE MNIST DATABASE”
Work done so far

➜ Attending course and assignments on ML by Andrew NG on


coursera.
➜Read and studied first four chapters on Neural Networks and Deep

Learning by Michael Nielsen and attained few basic knowledge


➜Learned and implemented basics of python, git and lua to abasic
level
➜Installed and implemented torch and loaded MNIST data.

➜Understood, tried and practiced MNIST tutorial provided by Andrea

Ferretti on RNDuja Blog.


Need to do

➜ Train vexo with data found on


http://www.ee.surrey.ac.uk/CVSSP/demos/chars74k

➜Use a better data set on


https://lvdmatten.github.io/software/code/wride
.tar.gz/

➜ Study convolutional neural networks and their


implementation on
http://cs231n.github.io/convolutional-networks/
➜Implement convolutional neural network architecture on
the old data set after filtering garbage data.

➜Use character segmentation code on MATLAB by Diego


Barragan, Technical University of Loja, Ecuador, available at
http://www.mathworks.com/matlabcentral/fileexchange/2
2922-image-segmentation---extractionfacilitating .

➜Used graph plotting tools to show graphs of loss vs time


and accuracy vs time.
The resulting system will
be a subset of a complex
OCR or ICR system

We expect a possible
future extensions of this
work
Reference

➜1st four chapters on Neural Network

and deep learning


➜www.ee.surrey.ac.uk

➜www.lvdmatten.github.io
Thanks!
Any questions?
CREDITS

Athul K S
Pranav M
Shino C S
V V Gokul

Вам также может понравиться