CHAPTER 1 INTRODUCTION

Digital images are widely used in computer applications. Uncompressed digital images require considerable storage capacity and transmission bandwidth. Efficient image compression solutions are becoming more critical with the recent growth of data intensive, multimedia based web applications. Data compression is the process of converting data files into smaller files for efficiency of storage and transmission. As one of the enabling technologies of the multimedia revolution, data compression is a key to rapid progress being made in information technology. It would not be practical to put images, audio, and video alone on websites without compression. Data compression algorithms are used in those standards to reduce the number of bits required to represent an image or a video sequence. Compression is the process of representing information in a compact form. Data compression treats information in digital form as binary numbers represented by bytes of data with very large data sets. Compression is a necessary and essential method for creating image files with manageable and transmittable sizes. In order to be useful, a compression algorithm has a corresponding decompression algorithm that, given the compressed file, reproduces the original file. There have been many types of compression algorithms developed. These algorithms fall into two broad types, lossless algorithms and lossy algorithms. A lossless algorithm reproduces the original exactly. A lossy algorithm, as its name implies, loses some data. Data loss may be unacceptable in many applications. Depending on the quality required of the reconstructed image, varying amounts of loss of information can be accepted. Video and Audio files are very large. There is need to compress an image to utilize the bandwidth effectively. Compression becomes part of the representation or coding scheme which have become popular audio, image and video formats. Compression basically employs redundancy in the data. Video and audio compression techniques are most suited to this form of lossy compression. Lossy compression use source encoding techniques that includes Huffman encoding. 1.1 BACKGROUND WORK
1

In certain applications like video conferencing and television, large amount of images are transmitted through internet. It necessitates an image compression technique to efficiently utilize the bandwidth. Various lossy compression techniques using Discrete Wavelet Transform are listed. Ability of these techniques is dependent on the selection of transforms. The performance is measured in terms of compression ratio. An extensive survey on various techniques used for Image compression using DWT are listed. 1.1.1 Review of DWT based image compression techniques Henry Y.H Chuang et al (1995) implemented VLSI architecture for fast 2D discrete orthonormal wavelet transform. They presented a parallel pipelined VLSI array architecture for 2D dyadic separable DWT. The 2D data array is partitioned into non-overlapping groups of rows. All rows in a partition are processed in parallel, and consecutive partitions are pipelined. Moreover, multiple wavelet levels are computed in the same pipeline, and multiple DWT problems can be pipelined also. The whole computation requires a single scan of the image data array. Thus, it is suitable for on-line real-time applications. For an NN image, an m-level DWT can be computed in N22q+2m1(2p+3) time units on a processor costing no more than 83(Cm+Cu)p(p+2q) , where q is the partition size, p is the length of corresponding 1D DWT filters, C
m

and C

are the costs of a parallel

multiplier and a parallel adder respectively, and a time unit is the time for a multiplication and an addition. For q=N m, the computing time reduces to N2. They found that when a large number of DWT problems are pipelined, the computing time is about N22q per problem. Chu Yu et al (1999) implemented a design of an efficient VLSI architecture for 2-D Discrete Wavelet Transforms. They presented VLSI architecture for the separable two-dimensional Discrete Wavelet Transform (DWT) decomposition. Using a computation-schedule table, they showed how the proposed separable architecture uses only a minimal number of filters to generate all levels of DWT computations in real time. For the computation of an N x N 2-D DWT with a filter length L, this architecture spends around N2 clock cycles, and requires 2NL-2N storage unit, 3L multipliers, as well as 3(L-1) adders. This architecture utilized a
2

computation-schedule

table

to

accomplish

all

the

resolution

levels

of

computations. Since this architecture has a low latency, a low hardware cost, and ability to process 2-D digital signals in real-time, it can be applied very well to the codec implementation for various video image processing standards, such as MPEG-4 and JPEG-2000. They found that For a video signal processing, the computation time of the architecture per picture spends about N 2 only, which meets the real-time requirement of many 2-D signal-processing applications. Ching-Hsien Chang et al (2000) presented an efficient VLSI architectures for fast computation of the Discrete Fourier Transform and its inverse. They have developed two new VLSI architectures for computing the point discrete Fourier transform (DFT) and its inverse (IDFT) based on a radix-2 fast algorithm, where is a power of two. First they have presented a linear systolic array that requires log2 complex multipliers and is able to provide a throughput of one transform sample per clock cycle. Compared with other related systolic designs based on direct computation or a radix-2 fast algorithm, the proposed one has the same throughput performance but involves less hardware complexity. This design is suitable for high-speed real-time applications, but it would not be easily realized in a single chip when gets large. It has been verified that the proposed memorybased architecture that can be realized in a single chip with throughput performance up to several Mega Transform samples per second for a transform length as high as several thousand. They found that Such area-time performance makes this design very competitive for use in long-length DFT applications, such as asymmetric digital subscriber lines (ADSL) and orthogonal frequency division multiplexing (OFDM) systems. Francescomaria Marino et al (2001) implemented two fast architectures for the direct 2-D Discrete Wavelet Transform. They have presented two architectures for the direct twodimensional (2-D) discrete wavelet transform (DWT). The first architecture is based on a modified recursive pyramid algorithm (MRPA) and performs a nonstandard decomposition (i.e., Mallats tree) of an image in approximately 2 2 3 clock cycles (ccs). They found that this consistently speeds up other known architectures that commonly need approximately 2 ccs.

Furthermore, the proposed architecture is simpler than others in terms of hardware complexity. Subsequently, they showed how symmetric/antisymmetric properties of linear-phase wavelet filter bases can be exploited in order to further reduce the VLSI area. This is used to design a second architecture that provides one processing unit for each level of decomposition (pipelined approach) and performs a decomposition in approximately 2 2 ccs. Jrg Ritter et al (2001) implemented a pipelined architecture for partitioned DWT based lossy image compression using FPGAs. They have presented a promising FPGA based hardware architecture for DWT in case of lossy and lossless image compression. The FPGA implementation proposed achieves its computational power by using the partitioned approach and taking advantage of pipelining. Currently, an EZT algorithm based on this 2D-DWT architecture is implemented within the scope of a masters thesis at our institute. They investigated the applicability of the approach presented for video compression, in particular to motion estimation. The partitioned approach is suitable for being applied to this area because motion seems to be usually localized through the sub images used. Furthermore, they have been looking for methods allowing the partitioned approach to result in higher compaction ratios than the non-partitioned algorithms. One point of departure is to classify the sub images with respect to similarity, to compress only one representative JC of each class C, and to represent the remaining sub images I of class C by the image I JC which specifies the differences between sub image I and its representative JC. Another point of departure is to exploit similarities between the trees of the multi-scale representation. Abdullah AlMuhit et al (2004) implemented VLSI implementation of Discrete Wavelet Transform (DWT) for image compression. They presented an approach towards VLSI implementation of the Discrete Wavelet Transform (DWT) for image compression. The design follows the JPEG2000 standard and can be used for both lossy and lossless compression. In order to reduce complexities of the design, linear algebra view of DWT and IDWT has been used. This design can be used for image compression in a robotic system. They developed an improved and

efficient Discrete Wavelet Transform algorithm and tested in MATLAB. Then the algorithm has been modified for implementation in VHDL. Jose Oliver et al (2006) implemented a Huffman coding of wavelet lower trees for very fast image compression. They have presented a very fast variation of the Lower-Tree Wavelet (LTW) image encoder. LTW is a fast non-embedded encoder with state-of-the-art compression efficiency, which employs a tree structure as a fast method of coding coefficients, being faster than other encoders like SPIHT or JPEG 2000.They have described the coding algorithm that all the wavelet subbands are scanned in 2x2 blocks of coefficients, from the first level to the Nth. They showed that this encoder is more efficient than other very fast wavelet encoders, like the recently proposed PROGRESS (which is surpassed in up to 0.5 dB), and faster than them (from 4 to 9 times in coding). Compared with the JPEG 2000 reference software, the encoder is from 18 to 38 times faster, while PSNR is similar at low bit-rates, and about 0.5 lower at high bit-rates. A very fast version of the lower-tree wavelet encoder using Huffman coding and other strategies have been presented to reduce execution time. The loss of coding efficiency is compensated by the reduction of execution time. They found that the encoder can be considered one of the fastest wavelet-based image encoders, and therefore it is good for real-time interactive multimedia communications, allowing simple implementations both in hardware and software. Barbara Penna et al (2007) implemented transform coding techniques for lossy hyper spectral data compression. Transform-based lossy compression has a huge potential for hyperspectral data reduction. Hyperspectral data are 3-D, and the nature of their correlation is different in each dimension. These calls for a careful design of the 3-D transform to be used for compression. They have investigated the transform design and rate allocation stage for lossy compression of hyper spectral data. First, they selected a set of 3-D transforms, obtained by combining in various ways wavelets, wavelet packets, the discrete cosine transform, and the KarhunenLove transform (KLT), and evaluate the coding efficiency of these combinations. Second, they proposed a low-complexity version of the KLT, in which complexity and performance can be balanced in a scalable way, allowing one to design the transform that better matches a specific
5

application. Third, they integrated this, as well as other existing transforms, in the framework of Part 2 of the Joint Photographic Experts Group (JPEG) 2000 standard, taking advantage of the high coding efficiency of JPEG 2000, and exploiting the interoperability of an international standard. An evaluation framework based on both reconstruction fidelity and impact on image exploitation has been introduced, and evaluate the proposed algorithm by applying this framework to AVIRIS scenes. They found that the scheme based on the proposed low-complexity KLT significantly outperforms previous schemes as to ratedistortion performance. Mukul Shirvaikar et al (2008) implemented VHDL implementation of wavelet packet transforms using SIMULINK tools. The real-time implementation of the Discrete Wavelet Transform (DWT) is a current area of research as it is one of the most time consuming steps in the JPEG2000 standard. The standard implements two different wavelet transforms: irreversible and reversible Daubechies. The former is a lossy transform, whereas the latter is a lossless transform. Many current JPEG2000 implementations are software based and not efficient enough to meet real-time deadlines. Field Programmable Gate Arrays (FPGAs) are revolutionizing image and signal processing. Many major FPGA vendors like Altera and Xilinx have recently developed SIMULINK tools to support their FPGAs. These tools are intended to provide a seamless path from system-level algorithm design to FPGA implementation. They investigated FPGA implementation of 2-D lifting-based Daubechies 9/7 and Daubechies 5/3 transforms using a Matlab/Simulink tool that generates synthesizable VHSIC Hardware Description Language (VHDL) code. They found that the feasibility of this approach for real time image processing by comparing the performance of the high-level toolbox with a handwritten VHDL implementation. Chao Cheng et al (2008) implemented high-speed VLSI implementation of 2-D Discrete Wavelet Transform. They presented a systematic high-speed VLSI implementation of the discrete wavelet transform (DWT) based on hardwareefficient parallel FIR filter structures. High-speed 2-D DWT with computation time as low as 2 12 can be easily achieved for an image with controlled increase of hardware cost. Hardware efficient parallel FIR filter structures are developed to
6

speed up the processing speed of 2-D DWT and to control the increase of hardware cost at the same time. They found that this design can be easily extended to achieve higher processing speed than the given highest processing speed with computing time and it is suitable for high-speed VLSI implementation of 2-D DWT because of its regular structure, simple control and 100% hardware utilization for continuous images. A. Mansouri et al (2009) implemented an efficient VLSI architecture and FPGA implementation of high-speed and low power 2-D DWT for (9, 7) wavelet Filter. They presented an efficient VLSI architecture of a high speed, low power 2D Discrete Wavelet Transform computing. This architecture, based on new and fast lifting scheme approach for (9, 7) filter in DWT, reduces the hardware complexity and memory accesses and this verified by simulation and successfully implemented in a Cyclone II and Stratix III FPGAs, and the estimated frequency of operation is 350 MHz. They found that the resulting computing rate is up to 48 frames (4096x2160) per second with 24 bpp. The architecture has regular structure, simple control flow, small embedded buffers and low power consumption. Thus, it is very suitable for new generation image compression systems, such as JPEG2000. Also they found that the advantages of the proposed architecture are saving embedded memories, fast computing time, low power consumption, and low control complexity. Chengjun Zhang et al (2010) presented a pipeline VLSI architecture for high-speed computation of the 1-D Discrete Wavelet Transform. They developed a scheme for the design of a high-speed pipeline VLSI architecture for the computation of the 1-D discrete wavelet transform (DWT). The main focus of the scheme is on reducing the number and period of clock cycles for the DWT computation with little or no overhead on the hardware resources by maximizing the inter- and intra stage parallelisms of the pipeline. The inter stage parallelism is enhanced by optimally mapping the computational load associated with the various DWT decomposition levels to the stages of the pipeline and by synchronizing their operations. The intra stage parallelism is enhanced by decomposing the filtering operation equally into two subtasks that can be performed independently in parallel and by optimally organizing the bitwise
7

operations for performing each subtask so that the delay of the critical data path from a partial-product bit to a bit of the output sample for the filtering operation is minimized. They showed that an architecture designed based on the proposed scheme requires a smaller number of clock cycles compared to that of the architectures employing comparable hardware resources. In fact, the requirement on the hardware resources of the architecture designed by using the proposed scheme also gets improved due to a smaller number of registers that need to be employed. Based on the proposed scheme, a specific example of designing an architecture for the DWT computation is considered. Finally, they found that the principle of maximizing the inter- and intra stage parallelisms for the design of architecture for the 1-D DWT computation is extendable to that for the 2-D DWT computation. Rohit Arora et al (2011) implemented an algorithm for image compression using 2D wavelet transform. The standard image MASK is analyzed using wavelet techniques for image compression. The transformations used for wavelet transformations are Haar, Symlet, Coiflet and db4. The results in terms of compression ratio and MSE show that the Haar transformation has the best performance. The Haar transform is then used to compress the image up to four levels. They sown that the reconstructed image after every level of compression. They found that DWT has potential application in compression problem and the use of Haar transformation for Wavelet Transformation provides the best results. Chengjun Zhang et al (2011) implemented a pipeline VLSI architecture for fast computation of the 2-D Discrete Wavelet Transform. They developed, a scheme for the design of a high-speed 3-stage pipeline VLSI architecture for the computation of the 2-D discrete wavelet transform (DWT). The main focus in the development of the architecture is on providing a high operating frequency and a small number of clock cycles along with an efficient hardware utilization by maximizing the inter-stage and intra-stage computational parallelism for the pipeline. The inter-stage parallelism is enhanced by optimally mapping the computational task of multi decomposition levels to the stages of the pipeline and synchronizing their operations. The intra-stage parallelism is enhanced by dividing the 2-D filtering operation into four subtasks that can be performed independently
8

in parallel and minimizing the delay of the critical path of bit-wise adder networks for performing the filtering operation. To validate this scheme, a circuit is designed, simulated, and implemented in FPGA for the 2-D DWT computation. They showed that the circuit is capable of operating with a maximum clock frequency of 134MHz and processing 1022 frames of size 512 512 per second with this operating frequency. They showed that the performance in terms of the processing speed of the architecture designed based on the proposed scheme is superior to those of the architectures designed using other existing schemes, and it has similar or lower hardware consumption. Finally, they found that the architecture designed is scalable in that its processing speed can be adjusted upward or downward by changing the number of MAC cells in each of the processing units by a factor equal to that of the reduction required in the processing speed. V. V. Sunil Kumar et al (2012) have developed image compression techniques by using wavelet transform. This is associated with a certain type of compression techniques by using wavelet transforms. Wavelets are used to characterize a complex pattern as a series of simple patterns and coefficients that, when multiplied and summed, reproduce the original pattern. The data compression schemes can be divided into lossless and lossy compression. Lossy compression generally provides much higher compression than lossless compression. Wavelets are a class of functions used to localize a given signal in both space and scaling domains.A Minimage was originally created to test one type of wavelet and the additional functionality was added to image to support other wavelet types and the EZW(Embedded Zerotree Wavelet) coding algorithm was implemented to achieve better compression results. 1.2 MOTIVATION Various image compression techniques were developed and stopped in simulation level itself. In order to implement an image compression technique on FPGA, it is necessary to develop VHDL code. This motivates to propose an efficient image compression technique based on Discrete Wavelet Transform and Huffman encoding using VHDL.

1.3 PROBLEM DEFINITION In this proposed work, aims at compressing an image using 2-d Discrete Wavelet Transform and Huffman encoding to increase the compression ratio and reduce the Data Redundancy. Discrete wavelet transform and Huffman encoding will be implemented using VHDL. The performance will be evaluated in terms of compression ratio for five different images. 1.4 OBJECTIVES
To acquire the images. To perform Discrete Wavelet Transform on the images. To perform Huffman Encoding for the approximation coefficients. To measure the compression ratio for different images.

1.5 SCOPE OF THE WORK The main scope of the project is to compress an image to utilize the bandwidth efficiently for certain applications like Television and Video Conferencing. Video and Audio files are very large. There is a need to compress an image to utilize the bandwidth effectively. Compression becomes part of the representation or coding scheme which have become popular audio, image and video formats. An efficient image compression technique that uses Huffman encoding for the approximation coefficients obtained from Discrete Wavelet Transform is used during the transmission of images through an internet. This compression technique based on Discrete Wavelet Transform and Huffman encoding is proposed to increase compression ratio and reduce data redundancy. This proposed work is highly suitable for an applications like video conferencing.

CHAPTER 2 OVERVIEW OF DWT AND HUFFMAN ENCODING

10

The wavelet transform (WT) provides an alternative approach to signal processing, especially suited for the analysis of spatial and spectral locality. In recent years, there has been a number of studies on wavelet transforms for signal analysis and synthesis. Generally, the 2- D DWT is frequently applied in image and video processing. The discrete wavelet transform (DWT) has been widely used in audio and image processing, digital communications and other application fields. This computation transform has been widely implemented in very-largescale integration (VLSI) because of the real-time requirement. Wavelets are functions defined over a finite interval. The basic idea of the wavelet transform is to represent an arbitrary function as a linear combination of a set of such wavelets or basis functions. These basis functions are obtained from a single prototype wavelet called the mother wavelet by dilations (scaling) and translations (shifts). The purpose of wavelet transform is to change the data from time-space domain to time-frequency domain which makes better compression results. The fundamental idea behind wavelets is to analyze the signal at different scales or resolutions, which is called multiresolution. Huffman coding is based on the frequency of occurrence of a data item (pixel in images). The principle is to use a lower number of bits to encode the data that occurs more frequently. Codes are stored in a code Book which may be constructed for each image or a set of images. In all cases the code book plus encoded data must be transmitted to enable decoding. 2.1 OVERVIEW OF DWT Wavelets are a class of functions used to localize a given signal in both space and scaling domains. A family of wavelets can be constructed from a mother wavelet. Compared to Windowed Fourier analysis, a mother wavelet is stretched or compressed to change the size of the window. In this way, big wavelets give an approximate image of the signal, while smaller and smaller wavelets zoom in on details. Therefore, wavelets automatically adapt to both the high-frequency and the low-frequency components of a signal by different sizes of windows. Any small change in the wavelet representation produces a correspondingly small change in the original signal, which means local mistakes will not influence the entire transform. The wavelet transform is suited for non
11

stationary signals, such as very brief signals and signals with interesting components at different scales. 2.1.1 Wavelet Decomposition There are several ways wavelet transforms can decompose a signal into various sub bands. These include uniform decomposition, octave-band decomposition, and adaptive or wavelet-packet decomposition. Out of these, octave-band decomposition is the most widely used. The procedure is as follows: wavelet has two functions wavelet and scaling function. They are such that there are half the frequencies between them. They act like a low pass filter and a high pass filter. The decomposition of the signal into different frequency bands is simply obtained by successive high pass and low pass filtering of the time domain signal. This filter pair is called the analysis filter pair. First, the low pass filter is applied for each row of data, thereby getting the low frequency components of the row. But since the low pass filter is a half band filter, the output data contains frequencies only in the first half of the original frequency range. By Shannon's Sampling Theorem, they can be subsampled by two, so that the output data now contains only half the original number of samples. Now, the high pass filter is applied for the same row of data, and similarly the high pass components are separated. In DWT, an image can be analysed by passing it through an analysis filter bank followed by decimation operation. An analysis filter bank consists of a low-pass and high-pass filter at each decomposition stage. When the signal passes through these filters, it splits into two bands. the low-pass filter, which corresponds to an averaging operation, extracts the coarse information of the signal. The high-pass filter which corresponds to a differencing operation, extracts the detail information of the signal. The output of the filtering operation is then decimated by two. A twodimensional transform is accomplished by performing two separate one dimensional transforms. This is shown in Fig 2.1.

12

Figure 2.1 Decomposition of an image This is a non-uniform band splitting method that decomposes the lower frequency part into narrower bands and the high-pass output at each level is left without any further decomposition. This procedure is done for all rows. Next, the filtering is done for each column of the intermediate data. The resulting twodimensional array of coefficients contains four bands of data, each labelled as LL (low-low), HL (high-low), LH (low-high) and HH (high-high). The LL band can be decomposed once again in the same manner, thereby producing even more sub bands. 2.1.2 Wavelet Transform using Haar Wavelets Image transforms are very important in digital processing they allow to accomplish less with more. For example the Fourier Transform may be used to effectively compute convolutions of images1 or the Discrete Cosine Transform may be used to significantly decrease space occupied by images without noticeable quality loss. Wavelet Transform (WT) is a relatively new concept as a whole, even it though it incorporates some of the transforms, which have been know for long time. Haar wavelet is a simplest wavelet and the wavelet coefficients are used with the scaling function.

13

2.1.2.1 Algorithm of Discrete Wavelet Transform Discrete wavelet Transform is given by

The following algorithm is also known as subband coding. Assume the highest frequency component that exists in a signal will be radians. To provide sufficient information about the signal without its loss due to sampling, the signal is sampled at Nyquists rate (which is twice the maximum frequency that exists in the signal). Next, the signal is passed through two filters one half band low pass and one half band high band filter. The output of the high pass filter is preserved. After passing the signal through a half band low pass filter, half of the samples can be eliminated according to the Nyquists rule, since the signal now has a highest frequency of /2 radians instead of radians Simply discarding every other sample will subsample the signal by two, and the signal will then have half the number of points. The scale of the signal is now doubled. Note that the low pass filtering removes the high frequency information, but leaves the scale unchanged. Only the sub sampling process changes the scale. Resolution, on the other hand, is related to the amount of information in the signal, and therefore, it is affected by the filtering operations. Half band low pass filtering removes half of the frequencies, which can be interpreted as losing half of the information. Therefore, the resolution is halved after the filtering operation. Note, however, the sub sampling operation after filtering does not affect the resolution, since removing half of the spectral components from the signal makes half the number of samples redundant anyway. Half the samples can be discarded without any loss of information. In summary, the low pass filtering halves the resolution, but leaves the scale unchanged. The signal is then sub sampled by 2 since half of the number of samples is redundant. This doubles the scale.

14

Figure 2.2 Subband coding

The transformation x = (x1, . . . , xN) (s | d) = (s1, . . . , sN/2 | d1, . . . , dN/2) is called the Discrete Haar Wavelet Transformation. 2.1.2.2 Formulation for Haar wavelet Transform Digital images require large amounts of memory to store and, when retrieved from the internet, can take a considerable amount of time to download. The Haar wavelet transform provides a method of compressing image data so that it takes up less memory. A 256 x 256 pixel gray scale image is stored as a 256 x 256 matrix, with each element of the matrix being a number ranging from zero (for black) to some positive whole number (for white). A 256 x 256 color image is stored as three 256 x 256 matrices (One each for the colors red, green, and blue). This matrix can be used and some linear algebra to maximize compression.

The Haar wavelet is the simplest wavelet.

15

1 0 < x <1/2 wH(x) = -1 < x < 1 0 otherwise The mother wavelet obviously satisfies the two wavelet requirements, as it is both local and oscillatory. The picture below shows the shapes of Haar wavelets for various scales and translations.

Figure 2.3 Shape of the Haar Wavelets Wavelet Transform is a type of signal representation that can give the frequency content of the signal at a particular instant of time or spatial location. The Haar wavelet transform decomposes the image into different sub-band images, It splits component into numerous frequency bands called sub-bands. They are LL, LH, HL, and HH sub-bands. A high-frequency sub-band contains the edge information of input image and LL sub-band contains the clear information about the image. An outstanding property of the Haar functions is that except function haar(0, t), the ith Haar function can be generated by the restriction of the (j 1)th function to be half of the interval where it is different from zero, by multiplication with p2 and scaling over the interval [0, 1]. These properties give considerable interest of the Haar function, since they closely relate them to the wavelet theory. In this setting, the first two Haar functions are called the global functions, while all the others are denoted as the local functions. Hence, the Haar function, which is an odd rectangular pulse pair, is the simplest and oldest wavelet. The motivation for using the discrete wavelet transform is to obtain information that is more
16

discriminating by providing a different resolution at different parts of the time frequency plane. The wavelet transforms allow the partitioning of the timefrequency domain into non uniform tiles in connection with the timespectral contents of the signal. The wavelet methods are strongly connected with classical basis of the Haar functions; scaling and dilation of a basic wavelet can generate the basis Haar functions. The Haar wavelet transform decomposes the image into different sub-band images, It splits component into numerous frequency bands called sub-bands. They are LL, LH, HL, and HH sub-bands. A high-frequency subband contains the edge information of input image and LL sub-band contains the clear information about the image. consider the Filter h=(h0,h1) y=h*x yn=1/2 xn + 1/2 xn-1 2.1.2.3. Point Haar Wavelet(oldest & simplest) g[n] = 1/2 for n = 1, 0 g[n] = 0 otherwise h[0] = 1/2, h[1] = 1/2, h[n] = 0 otherwise

2.1.2.4 2-Dimension(analysis) of Discrete wavelet transform

17

In DWT, an image can be analysed by passing it through an analysis filter bank followed by decimation operation. An analysis filter bank consists of a lowpass and high-pass filter at each decomposition stage. When the signal passes through these filters, it splits into two bands. the low-pass filter, which corresponds to an averaging operation, extracts the coarse information of the signal. The highpass filter which corresponds to a differencing operation, extracts the detail information of the signal. The output of the filtering operation is then decimated by two. A two-dimensional transform is accomplished by performing two separate one dimensional transforms. First, an image is filtered along the row and decimated by two. It is then followed by filtering the sub image along the column and decimated by two. This operation splits an image into four bands namely LL,LH,HL and HH respectively. Further decomposition can be achieved by acting upon the LL subband successfully and the resultant image is split into multiple bands. This process is shown in Fig 2.4.

. Figure 2.4 Two Dimension analysis of Discrete wavelet transform

2.1.2.5 Working of Haar Wavelet Transform

18

 The Haar wavelet uses a method of manipulating matrices called averaging and differencing. If y is a row of an image matrix, when it is averaged and differenced will get the following results. y 448 768 704 640 1280 1408 32 32 32 1600 64 64 64 1600 0 0 0

y1 608 672 1344 1600 160 y2 640 1472 32 128 160 y3 1056 416 32 128 160

Table 2.1 Average and Difference Blue = Approximation Coefficients Red = Detail Coefficients Using Matrix multiplication, but Y1=yA1 where A1 is the Matrix

1/2 1/2 A1= 0 0 0 0 0 0 and Y2=y1A2

0 0 1/2 1/2 0 0 0 0

0 0 0 0 1/2 1/2 0 0

0 0 0 0 0 0 1/2 1/2

1/2 -1/2 0 0 0 0 0 0

0 0 1/2 -1/2 0 0 0 0

0 0 0 0 1/2 -1/2 0 0

0 0 0 0 0 0 1/2 -1/2

where A2 is the Matrix

19

1/2 1/2 0 0 A2= 0 0 0 0

0 0 1/2 1/2 0 0 0 0

1/2 -1/2 0 0 0 0 0 0

0 0 1/2 -1/2 0 0 0 0

0 0 0 0 1 0 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1

and y3=y2A3 where A3 is the matrix

1/2 1/2 0 A3= 0 0 0 0 0

1/2 -1/2 0 0 0 0 0 0

0 0 1 0 0 0 0 0

0 0 0 1 0 0 0 0

0 0 0 0 1 0 0 0

0 0 0 0 0 1 0 0

0 0 0 0 0 0 1 0

0 0 0 0 0 0 0 1

This set of operations can all be done in one step: y3 = yW

20

Where W is is the transformation matrix: W = A1 A2 A3 Also note that since each column of the Ai matrices that comprise W is orthogonal to every other, the matrices are invertible. Thus: W^-1=A3^-1 A2^-1 A1^-1 This means that original can get data back using the equation. y = y3W^-1 In general, it says that Q = I W, where Q is the row transformed matrix and I is the original image matrix. But, the Haar wavelet transformation does these transformations to each column of the image matrix, and then repeats them on each column of the matrix. This is done by multiplying I on the left by the transpose of W. This gives us our final equation for the row-and-column transformed matrix T. T = WT I W start with an 8 x 8 image represented by matrix P 576 P= 704 768 832 832 960 768 448 704 640 832 832 832 896 768 768 1152 1280 1344 1472 1536 1536 1156 1088 1344 1408 1536 1600 1216 1472 1472 1536 1600 1600 960 960 896 832 704 1344 1536 1536 1600 1536 1216 1536 1600 1536 1536 1088 1600 1600 1600 1536 832 640 1280 1472 1600 1600 1280 1408 1600 1600

Last row is our vector y.

First there is a need to average and difference the rows of matrix P. In order to get the row averaged matrix Q we simply multiply P on the right by matrix W, our transformation matrix. This yields:
21

1200 272 1185 288 1312 240 1272 280 Q= 1256 296 1272 312 1144 344 1056 416 0

288 225 272 160 128 32 16 32 112 32 128

64 96 16 16 32

64 64 64 32 0 0 0 160

0 32 0 0 32 32 0

34 32 32 192 128 32 96 0 32 64 0

48 32 128

96 0

Now average and difference the columns of Q to get row and column transformed matrix T by multiplying Q on the left by WT . This yields:

1212 306 30 -50 T= 82 8 20 8 44 36 10 38 8 20 8 36

146 90 -20 24 32 56 48 0

54 24 68 2 24 68 16 0 8 8 0 48 48 20 72 64 48 32 16

40 8 16 32 16 16 16 16

4 4 16 8 16 16 16 0

16 16 80

16 16

Now matrix T is ready to be compressed.

2.1.2.6 Approximation and Detail Coefficients

22

If there is a need to average and difference each row and column of matrix P only once (T1 = AT1 P A1), we would essentially divide our 8 x 8 matrix into four 4 x 4 sub-matrices. The resulting filtered outputs are approximation and detail coefficients. That are approximation, vertical detail, horizontal detail and diagonal detail subbands of the input image. These contains the clear and edge information about an input image. Approximation coefficients contains the information of an image. The detail coefficients left out during the Huffman encoding to efficiently compress an image and also to increase the quality of an image. The Haar wavelet transform decomposes the image into different sub-band images, It splits component into numerous frequency bands called sub-bands. They are LL, LH, HL, and HH subbands. A high-frequency sub-band contains the edge information of input image and approximation sub-band contains the clear information about the image. These coefficients are shown in Fig 2.5.

656 1169 1392 1552 16 15 48 16 816 1248 1520 1584 16 160 16 16 880 1040 1584 1552 16 T1= 16 47 16 96 48 48 80 80 16 112 16 16 80 0 16 16 688 752 1360 1600 80 16 16 16 16 0 16 32 80

16 48 49 16 16

16 16 16 16 16 16 16 16 0

23

Figure 2.5 Coefficients of DWT 2.1.2.7 Applications of Haar Transform Due to its low computing requirements, the Haar transform has been mainly used for pattern recognition and image processing . Hence, two dimensional signal and image processing is an area of efficient applications of Haar transforms due to their wavelet-like structure. In this area, it is usually reported that the simplest possible orthogonal wavelet system is generated from the Haar scaling function and wavelet. Moreover, wavelets are considered as a generalization of the Haar functions and transforms. Such a transform is also well suited in communication technology for data coding, multiplexing and digital filtering. For example, application of non-normalized Haar transform in a sequence division multiplex system is described. Bandwidth economy for multiplexed digital channels based on Haar transform is presented in. For real time applications, hardware-based fast Haar chips have been developed. Different generalizations of Haar functions and
24

transform are used in digital speech processing with applications in voice controlled computing devices and robotics. The advantages of computational and memory requirements of the Haar transform make it of a considerable interest to VLSI designers as well. Digital MOS ICs as a tool external to the circuit, but also as a possibility for a self-test strategy. The use of this set of CAD tools allowed the derivation of strategies for testing MOS circuits when memory states were encountered as a consequence of some type of faults. The advantage of using Haar functions instead of Walsh functions in CAD system based on spectra methods for some classes of Boolean.

2.2 OVERVIEW HUFFMAN ENCODING Huffman code is a Optimum prefix code developed by D. Huffman in a class assignment. Construction of Huffman codes is based on two ideas: In an optimum code, symbols with higher probability should have shorter code words In an optimum prefix code, the two symbols that occur least frequently will have the same length (otherwise, the truncation of the longer codeword to the same length still produce a decodable code) Huffman coding is a popular method for compressing data with variablelength codes. Given a set of data symbols (an alphabet) and their frequencies of occurrence (or, equivalently, their probabilities), the method constructs a set of variable-length code words with the shortest average length and assigns them to the symbols. Huffman coding serves as the basis for several applications implemented on popular platforms. Some programs use just the Huffman method, while others use it as one step in a multistep compression process. The Huffman method [Huffman 52] is somewhat similar to the ShannonFano method, proposed independently by Claude Shannon and Robert Fano in the late 1940s ([Shannon 48] and [Fano 49]). It generally produces better codes, and like the ShannonFano method, it produces the best variable-length codes when the probabilities of the symbols are negative powers of 2. The main difference between the two methods is that ShannonFano constructs its codes from top to bottom (and the bits of each codeword are constructed from left to right), while Huffman constructs a code tree
25

from the bottom up (and the bits of each codeword are constructed from right to left). Since its inception in 1952 by D. Huffman, the method has been the subject of intensive research in data compression. The long discussion in [Gilbert and Moore 59] proves that the Huffman code is a minimum-length code in the sense that no other encoding has a shorter average length. A much shorter proof of the same fact was discovered by Huffman himself [Motil 07]. An algebraic approach to constructing the Huffman code is introduced in [Karp 61]. In [Gallager 78], Robert Gallager shows that the redundancy of Huffman coding is at most p1 + 0.086 where p1 is the probability of the most-common symbol in the alphabet. The redundancy is the difference between the average Huffman codeword length and the entropy. Given a large alphabet, such as the set of letters, digits and punctuation marks used by a natural language, the largest symbol probability is typically around 1520%, bringing the value of the quantity p1 + 0.086 to around 0.1. This means that Huffman codes are at most 0.1 bit longer (per symbol) than an ideal entropy encoder, such as arithmetic coding. The Huffman encoding algorithm starts by constructing a list of all the alphabet symbols in descending order of their probabilities. It then constructs, from the bottom up, a binary tree with a symbol at every leaf. This is done in steps, where at each step two symbols with the smallest probabilities are selected, added to the top of the partial tree, deleted from the list, and replaced with an auxiliary symbol representing the two original symbols. When the list is reduced to just one auxiliary symbol (representing the entire alphabet), the tree is complete. The tree is then traversed to determine the codewords of the symbols. Huffman coding is based on the frequency of occurrence of a data item (pixel in images). The principle is to use a lower number of bits to encode the data that occurs more frequently. Codes are stored in a Code Book which may be constructed for each image or a set of images. In all cases the code book plus encoded data must be transmitted to enable decoding.

26

2.2.1 Principle of Huffman codes Starting with two least probable symbols, g and d, of an alphabet A, if the codeword for g is [m]0, the codeword for d would be [m]1, where [m] is a string of 1s and 0s. Now, the two symbols can be combined into a group, which represents a new symbol y in the alphabet set. Five steps:
Find the gray-level probabilities for the image by finding the histogram Order the input probabilities (histogram magnitudes)from smallest to largest

Combine the smallest two by addition

Go to step2,until only two probabilities are left By working backward along the tree, generate code by alternating

assignment of 0 and 1. The symbol y has the probability P(g) + P(d). pattern [m] using the new alphabet set. Let A = {a1, , a5}, P(ai) = {0.2, 0.4, 0.2, 0.1, 0.1}. Recursively determine the bit

Table 2.2 Principle of Huffman codes

27

2.2.2 Efficiency of Huffman codes Redundancy the difference between the entropy and the average length of a code.

Table 2.3 Efficiency of Huffman codes The average codeword length for this code is l = 0.4 x 1 + 0.2 x 2 + 0.2 x 3 + 0.1 x 4 + 0.1 x 4 = 2.2 bits/symbol. The redundancy is 0.078 bits/symbol. For Huffman code, the redundancy is zero when the probabilities are negative powers of two. 2.2.3 Length of Huffman codes Given a sequence of positive integers {l1, l2, , lk} satisfies

there exists a uniquely decodable code whose codeword lengths are given by { l1, l2, , lk}. The optimal code for a source S has an average code length lavg with the following bounds:

where H(S) is the entropy of the source. The lower-bound can be obtained by showing that:

28

For the upper-bound, notice that given an alphabet {a1, a2, , ak}, and a set of codeword lengths.

the code satisfies the Kraft-McMillan inequality and has lavg < H(S) + 1. 2.2.4 Encoding Procedure Huffman coding is based on the frequency of occurrence of a data item (pixel in images). The principle is to use a lower number of bits to encode the data that occurs more frequently. Codes are stored in a Code Book which may be constructed for each image or a set of images. In all cases the code book plus encoded data must be transmitted to enable decoding. Huffman coding is a popular method for compressing data with variable-length codes. Given a set of data symbols (an alphabet) and their frequencies of occurrence (or, equivalently, their probabilities), the method constructs a set of variable-length code words with the shortest average length and assigns them to the symbols. An Encoding procedure for the Huffman encoding are as follows:
Find the gray-level probabilities for the image by finding the histogram

Order the input probabilities (histogram magnitudes)from smallest to largest Combine the smallest two by addition Go to step2,until only two probabilities are left
By working backward along the tree, generate code by alternating

assignment of 0 and 1.This is shown in Fig 2.6.

29

Figure 2.6 Encoding Procedure 2.2.5 Example of Huffman coding An example of Huffman encoding is shown if Fig 2.7. In such an alphabet, one symbol can be assigned the code 0 and the other code 1. The Huffman method cannot assign to any symbol a code shorter than one bit, so it cannot improve on this simple code. If the original data (the source) consists of individual bits, such as in the case of a bi-level (monochromatic) image, it is possible to combine several bits (perhaps four or eight) into a new symbol and pretend that the alphabet consists of these (16 or 256) symbols. The problem with this approach is that the original binary data may have certain statistical correlations between the bits, and some of these correlations would be lost when the bits are combined into symbols. When a typical bi-level image is digitized by scan lines, a pixel is more likely to be followed by an identical pixel than by the opposite one.

30

Figure 2.7 Huffman code for the 26-letter Alphabet Therefore have a file that can start with either a 0 or a 1 (each has 0.5 probability of being the first bit). A zero is more likely to be followed by another 0 and a 1 by another 1. Figure 2.7 is a finite-state machine illustrating this situation. If these bits are combined into, say, groups of eight, correlated by the original pixel probabilities. If the input data contains, e.g., the two adjacent groups 00011100 and 00001110, they will be encoded independently, ignoring the correlation between the last 0 of the first group and the first 0 of the next group. Selecting larger groups improves this situation but increases the number of groups, which implies more storage for the code table and longer time to calculate the table.

31

CHAPTER 3 DWT BASED IMAGE COMPRESSION TECHNIQUE FOR VIDEO CONFERENCING

A scheme of compression technique is used based on the combination of 2-D DWT/Huffman encoding is proposed to compress an input image by reducing the coding redundancy. Here, the Discrete wavelet transform and Huffman coding are implemented using VHDL. Input image size of 256x256 is read using MATLAB R2012 and image pixels are stored in Memory block module to perform Discrete Wavelet Transform. Here, the image compression technique is based on Discrete wavelet transform and Huffman coding. Approximation coefficients are obtained from DWT using VHDL source code. Huffman coding of approximation coefficients are obtained using VHDL and Compression ratio is measured. Simulation of DWT and Huffman Coding using Modelsim is performed. The block diagram of proposed work is shown in Fig 3.1. Input image 256x256
Image acquisition Memory block module Discrete Wavelet Transform with Haar module Huffman Encoding module

Compressed image in terms of compression ratio

compressed image

Figure 3.1 Block Diagram of proposed system

32

3.1 INPUT IMAGE AND MEMORY BLOCK An input image is red and it is converted into gray scale image. An image pixel values are created and stored in memory. This is show in figure 4.1.

Image acquisition

Convert an input image into gray scale image

Create text file containing pixel values

Store the image pixel values in memory block module Figure 3.2 Flowchart to read an input image and Memory block module 3.2 DISCRETE WAVELET TRANSFORM MODULE This module computes the wavelet transform coefficients of the input image pixels obtained from the memory module. In this band splitting method that decomposes the lower frequency part into narrower bands and the high-pass output at each level is left without any further decomposition. This procedure is done for all rows. Next, the filtering is done for each column of the intermediate data. The resulting two-dimensional array of coefficients contains four bands of data, each labelled as LL (low-low), HL (high-low), LH (low-high) and HH (highhigh). The Low pass and High pass coefficients are computed. This is shown in figure 3.2.
33

Start

Image acquisition

Apply horizontal decomposition

Apply vertical decomposition

DWT Coefficients

Stop Figure 3.3 Flowchart of DWT The Approximation coefficients are applied as an input to Huffman encoding module to code the data. The Huffman data for those coefficients are determined. Further decomposition can be achieved by acting upon the LL subband successfully and the resultant image is split into multiple bands. An input sequence is decomposed into low-pass and high-pass sub bands in this method of DWT with Haar wavelet. This Haar wavelet transform decomposes the image into different sub-band images, It splits component into numerous frequency bands called sub-bands. They are LL, LH, HL, and HH sub-bands. A high-frequency subband contains the edge information of input image and LL sub-band contains the clear information about the image.

34

3.3 HUFFMAN ENCODING MODULE The Huffman encoding of DWT output has been computed. The compressed values are viewed serially. Find the gray-level probabilities for the image by finding the histogram

Order the input probabilities (histogram magnitudes) from smallest to largest

Combine the smallest two by addition

No

Are only two probabilities are left? Yes By working backward along the tree, generate code by alternating assignment of 0 and 1

Figure 3.4 Flowchart of Huffman encoding module 3.3.1 Huffman Code

35

The reference codeword for the approximation coefficients along with codeword length are shown in Table 2.4.

Table 2.4 Huffman Code 3.4 COMPRESSED IMAGE The compressed image is displayed in terms of compression ratio. The compressed values are viewed serially. This s shown if figure 3.4. View the compressed values serially

Obtain the compression ratio

Measure the performance by obtaining compression ratio for different images

Figure 3.5 Flowchart for compressed image in terms of compression ratio

3.5 OVERALL FLOWCHART

36

An image is red and the pixels are stored as a text file. The approximation coefficients are obtained by implementing the DWT with haar. Then, the Huffman encoding of Approximation coefficients are described to find the compressed values. Compressed image is obtained in terms of Compression ratio.

Start Image acquisition

Perform DWT with Haar Wavelet and obtain the approximation ad detail coefficients

Perform Huffman Encoding of Approximation coefficients

Compressed image Stop

Figure 3.6 Flowchart of proposed system

CHAPTER 4 RESULTS AND DISCUSSION

37

4.1 INPUT IMAGES The Different colour images are given as an input for the DWT based imaged compression. The input images are shown below.

Figure 4.1. Input Image 1

Figure 4.2. Input Image 2

Figure 4.3. Input Image 3

Figure 4.4. Input Image 4

38

Figure 4.5. Input Image 5 4.2 GRAY SCALE IMAGES The gray scale images for the given input images are shown below.

Figure 4.6. Gray scale Image 1

Figure 4.7. Gray scale Image 2

Figure 4.8. Gray scale Image 3

Figure 4.9. Gray scale Image 4

Figure 4.10 Gray scale Image 5 4.3 SIMULATION OUTPUT FOR DISCRETE WAVELET TRANSFORM Discrete Wavelet Transform with Haar wavelet is simulated using VHDL. Memory bock module is used to store the image pixel values, then the pixels are read from memory through the data lines and wavelet coefficients are obtained. The LL, LH,HL,HH coefficients are calculated using wavelet decomposition. The simulation output of DWT for an image 1 is shown in Fig 4.11.

Figure 4.11 Simulation output of DWT for image 1

39

The DWT coefficients of an image 1 are LL(Low-Low), LH(Low-High), HL(High-Low), HH(High-High) are shown in Fig 4.12.

Figure 4.12 DWT coefficients for image 1

Figure 4.13 Simulation output of DWT for image 2

The DWT coefficients of an image 2 are shown in Fig 4.14.

Figure 4.14 DWT coefficients for image 2 Discrete Wavelet Transform with Haar wavelet is simulated using VHDL. Memory bock module is used to store the image pixel values, then the pixels are read from memory through the data lines and wavelet coefficients are obtained. The LL, LH,HL,HH coefficients are calculated using wavelet decomposition. The pixel values and the corresponding coefficients of those pixel values are shown in Fig 4.13. An address lines and data lines which are used in memory block are also specified. Whenever the clock pin is high the process will be initialised and the coefficients are determined.

Figure 4.15 Simulation output of DWT for image 3 Discrete Wavelet Transform with Haar wavelet is simulated using VHDL. Memory bock module is used to store the image pixel values, then the pixels are read from memory through the data lines and Wavelet coefficients are obtained. The LL, LH,HL,HH coefficients are calculated using wavelet decomposition. The pixel values and the corresponding coefficients of those pixel values are shown in Fig 4.15. An address lines and data lines which are used in memory block are also specified. Whenever the clock pin is high the process will be initialised and the
40

coefficients are determined. The DWT coefficients for an image 3 are shown in Fig 4.16.

Figure 4.16 DWT coefficients for image 3 Discrete Wavelet Transform with Haar wavelet is simulated using VHDL. Memory bock module is used to store the image pixel values, then the pixels are read from memory through the data lines and wavelet coefficients are obtained. The LL, LH,HL,HH Coefficients are calculated using wavelet decomposition. The Pixel values and the corresponding coefficients of those pixel values are shown in Fig 4.17. An address lines and data lines which are used in memory block are also specified. Whenever the clock pin is high the process will be initialised and the coefficients are determined. The DWT coefficients for an image 4 are shown in Fig 4.18.

Figure 4.17 Simulation output of DWT for image 4

Figure 4.18 DWT coefficients for image 4

Figure 4.19 Simulation output of DWT for image 5 Discrete Wavelet Transform with Haar wavelet is simulated using VHDL. Memory bock module is used to store the image pixel values, then the pixels are read from memory through the data lines and wavelet coefficients are obtained. The LL, LH,HL,HH coefficients are calculated using wavelet decomposition. The Pixel values and the corresponding coefficients of those pixel values are shown in Fig 4.19. An address lines and data lines which are used in memory block are also specified. Whenever the clock pin is high the process will be initialised and the coefficients are determined. The DWT coefficients for an image 5 are shown in Fig 4.20.

41

Figure 4.20 DWT coefficients for image 5 4.4 SIMULATION OUTPUT FOR HUFFMAN ENCODING The approximation coefficients are given as an input to Huffman data table to obtain the Huffman data for each coefficient. Huffman code variable has an input of codeword length and the compressed data is viewed as serially. Huffman code for the approximation coefficients of an image 1 is shown in Fig 4.21.

Figure 4.21 Simulation output of Huffman coding for image 1 The approximation coefficients are given as an input to Huffman data table to obtain the Huffman data for each coefficient. Huffman code variable has an input of codeword length and the compressed data is viewed as serially. Huffman code for the approximation coefficients of an image 2 is shown in Fig 4.22.

Figure 4.22 Simulation output for image 2 The approximation coefficients are given as an input to Huffman data table to obtain the Huffman data for each coefficient. Huffman code variable has an input of codeword length and the compressed data is viewed as serially. Huffman code for the approximation coefficients of an image 3 is shown in Fig 4.23.

Figure 4.23 Simulation output for image 3 The approximation coefficients are given as an input to Huffman data table to obtain the Huffman data for each coefficient. Huffman code variable has an input of codeword length and the compressed data is viewed as serially.

42

The Huffman codewords for the approximation coefficients of an image 4 is shown in Fig 4.24.

Figure 4.24 Simulation output for image 4

The approximation coefficients are given as an input to Huffman data table to obtain the Huffman data for each coefficient. Huffman code variable has an input of codeword length and the compressed data is viewed as serially. Huffman code for the approximation coefficients of an image 5 is shown in Fig 4.25.

Figure 4.25 Simulation output for image 5

4.5 COMPRESSED VALUES AND COMPRESSION RATIO The compressed values and compression ratio of an image 1 is shown Fig 4.26.

Figure 4.26 Compressed values and compression ratio for image 1 The compressed values and compression ratio of an image 2 is shown Fig 4.27.

Figure 4.27 Compressed values and compression ratio for image 2

The compressed values and compression ratio of an image 3 is shown Fig 4.28.
43

Figure 4.28 Compressed values and compression ratio for image 3 The compressed values and compression ratio of an image 4 is shown Fig 4.29.

Figure 4.29 Compressed values and compression ratio for image 4

The compressed values and compression ratio of an image 5 is shown Fig 4.30.The parallel compressed values are converted into serial and these are viewed serially by column wise. All the serial values are showing the corresponding Huffman code for the approximation coefficients.

Figure 4.30 Compressed values and compression ratio for image 5

The Compression ratio for five different images are compared and these ratio are shown in Table 5.1. Table 5.1 Comparison of Compression ratio for five different images

Images Desert Hydrangeas Lighthouse Penguins Tulips

Compression ratio 0.9478 0.9389 0.9532 0.9524 0.9536

44

The compression ratio for five different images are plotted as a graph. This is shown in Fig 4.31.

Figure 4.31 Flow graph of compression ratio for five different images

CHAPTER 5 CONCLUSION
In this work, an efficient compression technique that uses Huffman encoding for the approximation coefficients of the given images has been successfully developed. In the first stage, the discrete wavelet transform is performed using Haar wavelet filter coefficients. The DWT coefficients are approximation and detail coefficients. These coefficients are determined using wavelet decomposition. In the second stage, the Huffman encoding for the approximation coefficients are performed. The compressed values are viewed serially by converting the parallel values. The Huffman code for each coefficient is determined. The performance of the proposed technique is evaluated in terms of compression ratio for five different images. It is found that the compression ratio has decreased significantly as only approximation coefficients are considered as the same time the quality of image has been compromised as detail coefficients are left out. The proposed lossy compression technique is highly suitable for applications like video conferencing during the transmission of images through an internet but not that suitable for medical image transmission also the compression ratio can be further reduced by incorporating quantization of the approximation coefficients.

45

CHAPTER 6 BIBLIOGRAPHY

[1] Angelopoulou. M, Masselos. K, Cheung. P and Andreopoulos. Y (2007), A comparison of 2-D discrete wavelet transform computation schedules on FPGAs, in Proc. IEEE Int. Conf. Field Programmable Technology (FPT), Bangkok, Tailand, pp. 181188. [2] Averbuch. A, Lazar. D and Israeli. M (1996), Image compression using wavelet transform and multiresolution decomposition. IEEE Transactions on Image Processing, vol. 5, no. 1, pp. 4-15. [3] Chao Cheng (2008), Member, IEEE High-Speed VLSI Implementation of 2-D Discrete Wavelet Transform, IEEE Trans, Signal Processing, Vol. 56, No. 1.

[4] Chengjun Zhang, Chunyan Wang (2010), Senior Member, IEEE A Pipeline VLSI Architecture for High-Speed Computation of the 1-D Discrete Wavelet Transform, IEEE Transactions On Circuits and SystemsI: Regular Papers, Vol. 57, No. 10.

[5] Ching-Hsien Chang, Chin-Liang Wang (2000), Member, IEEE Efficient VLSI Architectures for Fast Computation of the Discrete Fourier Transform and Its Inverse, IEEE Transactions On Signal Processing, Vol. 48, No. 11.

[6] Chrysytis. C and Ortega. A (2000), Line-based, reduced memory, wavelet image compression, IEEE Trans. Circuits Syst. Video Technol., vol. 9, no. 3, pp. 378389.
46

[7] Grzesczak. A, Mandal. M. K and Panchanathan. S (1996), VLSI implementation of discrete wavelet transform, IEEE Trans. Very Large ScaleIntegr. (VLSI) Syst., vol. 4, no. 4, pp. 421433.

[8] Mallat. S (2009), A theory for multiresolution signal decomposition : The wavelet representation, IEEE Trans. Pattern Anal. and Machine Intell.

[9] Mansouri. A (2009) An Efficient VLSI Architecture and FPGA Implementation of High-Speed and Low Power 2-D DWT for (9, 7) Wavelet Filter, IJCSNS International Journal of Computer Science and Network Security, VOL.9 No.3.

[10] Raghunath. S and Aziz. S. M (2006), High speed area efficient multiresolution 2-D 9/7 filter DWT processor, in Proc. Int. Conf. Very Large Scale Integration (IFIP), vol. 1618, pp. 210215. [11] Rioul. O and Vetterli. M (1991), Wavelets and signal processing, IEEE Signal Processing Magazine, vol. 8, no.4, pp. 14-38. [12] Sunil kumar. V. V (2012) Image Compression Techniques by using Wavelet Transform, Journal of International Engineering and Applications ISSN 2224-5782 (print) ISSN 2225-0506 (online) Vol 2, No.5.

47