application of generalized representations for image compression application of generalized...

11
Application Application of Generalized of Generalized Representations Representations for Image Compression for Image Compression using Vector Quantization Technion - Israel Institute of Technology Department of Electrical Engineering The Image and Computer Vision Laboratory Sheingart Michael & Tseitlin Yuri Supervisor: Wajcer Daniel

Post on 19-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

ApplicationApplicationof Generalized Representationsof Generalized Representationsfor Image Compressionfor Image Compressionusing Vector Quantization

Technion - Israel Institute of TechnologyDepartment of Electrical EngineeringThe Image and Computer Vision Laboratory

Sheingart Michael & Tseitlin Yuri

Supervisor: Wajcer Daniel

Page 2: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

Project GoalsProject Goals

Checking an efficiency of the multifrequency representation used for image compressing and coding .

Implementing and simulation of the Wavelet Image Coder .

Comparation with another families of Image Coders .

Page 3: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

The general structureThe general structure

The Wavelet Transform .The Vector Quantization .The Entropy Coder is

implemented by Adaptive Arithmetic Coding .

Page 4: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

The Wavelet Coder SchemeThe Wavelet Coder Scheme

D esired R a te

T ra n sm ittedp ic tu re

T ra in in g p ic tu re

T ra in in g o f th e Q u an tiz e r

P ru n in g o f th e co d eb o o k tre e b y R O P A

R eco m p u te th e n ew p a rtitio n v ec to rb y q u an tiz a tio n o f th e tran sm itted p ic tu re

A d ap tiv e A rith m e tic C o d in g o fn ew p a rtitio n v ec to r

D eco d in g o f p a rtitio n v ec to r

T h e reco n stru cted p ic tu re

R eco n s tru c tin g o f th e tran sm ittedp ic tu re u s in g th e v ec to r q u an tiz e r co d eb o o k

T ran sm ittin g

Page 5: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

The reasons for using Wavelet The reasons for using Wavelet TransformTransform Majority of information

is stored in low pass filter and high pass filters are very sparse .

The lack of redundancy between the transform coefficients .

The transform is mathematically stable .

Page 6: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

The Vector QuantizationThe Vector Quantization

Computing the binary tree by Splitting Algorithm and training the codebooks by Max-Lloyd Algorithm .

Pruning the tree by ROPA Algorithm .

The Bit Allocation Algorithm to obtain the desired Rate .

Page 7: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

The Adaptive Arithmetic CoderThe Adaptive Arithmetic Coder

The coder is free from the statistical issues associated with the design of Huffman Code .

The coder adapts to varied statistic features of a source .

There is free error coding .

Page 8: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

Comparative performance of Comparative performance of the Vector Quantizerthe Vector Quantizer

The comparation between Entropy -Constrained Scalar Quantizer ( right ) and our Vector Quantizer ( left ) .

Page 9: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

Comparative performance of Comparative performance of the Wavelet Image Coderthe Wavelet Image Coder

The comparation between Embedded Zerotree Wavelet Coder ( right ) and our Wavelet Image Coder ( left ) .

Page 10: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

Performance of Wavelet Image Performance of Wavelet Image CoderCoder

1. Original picture2. Quantized picture with __Rate = 1 bpp3. Coded picture with __training codebook

2

1

3

Page 11: Application of Generalized Representations for Image Compression Application of Generalized Representations for Image Compression using Vector Quantization

ConclusionConclusion

The wavelet representation is powerful tool for image coding and compression .

The vector quantization using Splitting Algorithm yields a minimum distortion quantizer for fixed code word length vectors .

Any intermediate rate can be obtained . The quantizer is sub-optimal but easy

for training and using since it is tree-structured quantizer .