A New Image Compression Scheme Based on Locally Adaptive Coding

Slides:



Advertisements
Similar presentations
דורון דמרי מקסים גורביץ מנחה : דמיטרי פורמן Cached Vector Quantization.
Advertisements

Vector Quantization. 2 outline Introduction Two measurement : quality of image and bit rate Advantages of Vector Quantization over Scalar Quantization.
Fast vector quantization image coding by mean value predictive algorithm Authors: Yung-Gi Wu, Kuo-Lun Fan Source: Journal of Electronic Imaging 13(2),
Colored Watermarking Technology Based on Visual Cryptography Author: Hsien-Chu Wu, Chwei-Shyong Tsai, Shu-Chuan Huang Speaker: Shu-Chuan Huang Date: May.
1 An Efficient VQ-based Data Hiding Scheme Using Voronoi Clustering Authors:Ming-Ni Wu, Puu-An Juang, and Yu-Chiang Li.
1 Information Hiding Based on Search Order Coding for VQ Indices Source: Pattern Recognition Letters, Vol.25, 2004, pp.1253 – 1261 Authors: Chin-Chen Chang,
1 資訊隱藏技術之研究 The Study of Information Hiding Mechanisms 指導教授: Chang, Chin-Chen ( 張真誠 ) 研究生: Lu, Tzu-Chuen ( 呂慈純 ) Department of Computer Science and Information.
Palette Partition Based Data Hiding for Color Images Yu-Chiang Li, Piyu Tsai, Chih-Hung Lin, Hsiu-Lien Yeh, and Chien-Ting Huang Speaker : Yu-Chiang Li.
A Fast LBG Codebook Training Algorithm for Vector Quantization Presented by 蔡進義.
Advisor: Chang, Chin-Chen Student: Chen, Chang-Chu
基於(7,4)漢明碼的隱寫技術 Chair Professor Chin-Chen Chang (張真誠)
影像偽裝術 Dr. Chin-Chen Chang
Chair Professor Chin-Chen Chang Feng Chia University Jan. 2008
A Secret Information Hiding Scheme Based on Switching Tree Coding
Reporter :Liang, Min-Chen Date :2003/04/15
Data Mining and Its Applications to Image Processing
An Information Hiding Scheme Using Sudoku
Information Steganography Using Magic Matrix
Chapter 3 向量量化編碼法.
Advisor: Chin-Chen Chang1, 2
Chair Professor Chin-Chen Chang Feng Chia University Aug. 2008
Information Steganography Using Magic Matrix
A Color Image Hiding Scheme Based on SMVQ and Modulo Operator
Chair Professor Chin-Chen Chang Feng Chia University
Source :Journal of visual Communication and Image Representation
Chair Professor Chin-Chen Chang Feng Chia University
Advisor: Chin-Chen Chang1, 2 Student: Wen-Chuan Wu2
A Data Hiding Scheme Based Upon Block Truncation Coding
第七章 資訊隱藏 張真誠 國立中正大學資訊工程研究所.
Hiding Data in a Color Palette Image with Hybrid Strategies
A Study of Digital Image Coding and Retrieving Techniques
Embedding Secrets Using Magic Matrices
Advisor: Chin-Chen Chang1, 2 Student: Yi-Pei Hsieh2
An Innovative Steganographic Scheme Based on Vector Quantization
A Restricted Region-based Data-hiding Scheme
An Innovative Steganographic Scheme Based on Vector Quantization
第 四 章 VQ 加速運算與編碼表壓縮 4-.
Reversible Data Hiding Scheme Using Two Steganographic Images
Advisor:Prof. Chin-Chen Chang Student :Kuo-Nan Chen
Density-Based Image Vector Quantization Using a Genetic Algorithm
Chair Professor Chin-Chen Chang (張真誠) National Tsing Hua University
Information Steganography Using Magic Matrix
Dynamic embedding strategy of VQ-based information hiding approach
Chair Professor Chin-Chen Chang Feng Chia University
Hung, K. -L. and Chang, C. -C. , IEE Image and Signal Processing, vol
A Self-Reference Watermarking Scheme Based on Wet Paper Coding
A Color Image Hiding Scheme Based on SMVQ and Modulo Operator
Hiding Information in VQ Index Tables with Reversibility
Information Hiding and Its Applications
Chair Professor Chin-Chen Chang (張真誠) National Tsing Hua University
Chair Professor Chin-Chen Chang Feng Chia University
Zhe-Ming Lu, Chun-He Liu, Dian-Guo Xu, Sheng-He Sun,
Authors: Chin-Chen Chang, Yi-Hui Chen, and Chia-Chen Lin
第 十 章 隱像術.
A Virtual Image Cryptosystem Based upon Vector Quantization
Hung, K. -L. and Chang, C. -C. , IEE Image and Signal Processing, vol
A Robust and Recoverable Tamper Proofing Technique for Image Authentication Authors: Chin-Chen Chang & Kuo-Lung Hung Speaker : Chin-Chen Chang.
A Self-Reference Watermarking Scheme Based on Wet Paper Coding
De-clustering and Its Application to Steganography
A Data Hiding Scheme Based Upon Block Truncation Coding
Source: Pattern Recognition, Volume 40, Issue 2, February 2007, pp
Predictive Grayscale Image Coding Scheme Using VQ and BTC
資訊偽裝術 張真誠 講座教授 多媒體暨網路安全實驗室
Information Hiding Techniques Using Magic Matrix
Authors: Chin-Chen Chang, Yi-Hui Chen, and Chia-Chen Lin
Steganographic Systems for Secret Messages
Chair Professor Chin-Chen Chang Feng Chia University Jan. 2008
A New Image Compression Scheme Based on Locally Adaptive Coding
A Restricted Region-based Data-hiding Scheme
Hiding Information in VQ Index Tables with Reversibility
Presentation transcript:

A New Image Compression Scheme Based on Locally Adaptive Coding Chair Professor Chin-Chen Chang Feng Chia University

Introduction Image compression technique w h Image Index table Vector Quantization Encoder

Introduction (Cont.) Image compression technique w h Image Index table Vector Quantization Decoder

Introduction (Cont.) Histogram VQ Index table Image 207 246 : 218 223 … VQ Histogram Index table Image

Introduction (Cont.) Locally adaptive scheme Ex: a a b b c a c a b d c c e f b a List List a b a 0 1 List List List List … c b a a c b c a b a c b 002 012 102 002 012 102 002 012 102 002 012 102 0 a 1 0 0 b 1 0 0 c 1 10 1 01 1 01 … 0: the symbol does not exist in list 1: index existed in list Concept: Moving the Recently Used to the Front

The Proposed Method Image I Divide the temporary codebook into two Parts (frequently-used and infrequently-used) Refine the frequently-used part of temporary codebook Divide image into non-overlapping blocks Generate temporary codebook Image I Refine the infrequently-used part of temporary codebook generated codebook Output Compress I by referencing the generated codebook Compression code

The Proposed Method (Cont.) The proposed method includes two phases for codebook generation Temporary codebook generation Codebook refining

Temporary Codebook Generation The codebook generation is based on the concept of locally adaptive coding Temporary Codebook w Bi = {84, 170, 166, 167, 83, …, 168} G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} h G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} ∵ d(Bi , g2,1) = 2.24 TH = 25 ∴ add Bi into G2 and move G2 to be the first one Image Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} *d(•): Euclidean distance function

Temporary Codebook Generation (Cont.) G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}}

Temporary Codebook Generation (Cont.) w G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} Bj = {220, 212, 182, 179, 240, …, 246} h G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} Image Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} ∵ No any block satisfies d(Bj, gk) TH = 25 ∴ Create a new group Gn+1, add Bj into Gn+1, and move Gn+1 to be the first one

Temporary Codebook Generation (Cont.) G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} Gn+1 = { gn+1,1={220, 212, 182, 179, 240, …, 246}} G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} Gn+1 = { gn+1,1={220, 212, 182, 179, 240, …, 246}}

Codebook Refining ||Gn+1|| =1 Frequently-used ||G2|| =3 Gn ||G1|| =3 IF ||Gi|| > THF = 3 Then Add Gi into Frequently-used part Else add Gi into Infrequently-used part Temporary Codebook Gn+1 = { gn+1,1={220, 212, 182, 179, 240, …, 246}} ||Gn+1|| =1 G2 = { g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} Frequently-used ||G2|| =3 G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} Gn ||G1|| =3 G3 G3 = { g3,1={184, 83, 194, 170, 166, …, 110}, g3,2={180, 80, 194, 166, 168, …, 114}, g3,3={185, 85, 190, 172, 160, …, 120}, g3,4={191, 88, 197, 170, 162, …, 107}} ||G3|| =4 Infrequently-used Gn+1 Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} G2 ||Gn|| =4 G1

Codebook Refining (Cont.) For frequently-used part c3={185, 84, 194, 170, 164, …, 113} Compute Gi’s centroid cn={243, 85, 176, 211, 164, …, 127} G3 = { g3,1={84, 83, 94, 170, 166, …, 210}, g3,2={80, 80, 94, 166, 168, …, 214}, g3,3={85, 85, 90, 172, 160, …, 220}, g3,4={91, 88, 97, 170, 162, …, 207}} : ∵ d(c3, cn) = 32 THH = 35 ∴ Combine c3 and cn by c3’ = (c3 + cn)/2 Use the centroids to represent the codewords c3={185, 84, 194, 170, 164, …, 113} Gn = { gn,1={244, 83, 174, 210, 166, …, 127}, gn,2={243, 80, 175, 213, 168, …, 128}, gn,3={240, 85, 176, 209, 160, …, 126}, gn,4={244, 88, 178, 213, 162, …, 127}} cn={243, 85, 176, 211, 164, …, 127} c3’={214, 85, 185, 191, 164, …, 120} : Centroid of Gn

Codebook Refining (Cont.) For infrequently-used part c1={200, 96, 85, 167, 83, …, 131} c2={85, 170, 166, 169, 86, …, 168} Compute Gi’s centroid cn+1={220, 212, 182, 179, 240, …, 246} Gn+1 = {gn+1,1={220, 212, 182, 179, 240, …, 246}} cn+1={220, 212, 182, 179, 240, …, 246} Use the centroids to represent the codewords G2 = {g2,1={84, 170, 166, 167, 84, …, 170}, g2,2={86, 168, 165, 170, 90, …, 165}, g2,3={84, 170, 166, 167, 83, …, 168}} c2={85, 170, 166, 169, 86, …, 168} G1 = { g1,1={199, 96, 84, 167, 83, …, 130}, g1,2={201, 97, 90, 167, 85, …, 132}, g1,3={200, 95, 82, 167, 81, …, 131}} c1={200, 966, 85, 167, 83, …, 131}

Codebook Refining (Cont.) Frequently-used part C3’={214, 85, 185, 191, 164, …, 120} : ∵ d(c3’, c1) = 132.57 THL = 135 ∴ Delete c1 from Infrequently-used part Infrequently-used part c1={200, 96, 85, 167, 83, …, 131} c1={200, 96, 85, 167, 83, …, 131} c2={85, 170, 166, 169, 86, …, 168} c2={85, 170, 166, 169, 86, …, 168} cn+1={220, 212, 182, 179, 240, …, 246} cn+1={220, 212, 182, 179, 240, …, 246}

Codebook Refining (Cont.) Frequently-used part C3’={214, 85, 185, 191, 164, …, 120} C3’={214, 85, 185, 191, 164, …, 120} c2={85, 170, 166, 169, 86, …, 168} : : Infrequently-used part c2={85, 170, 166, 169, 86, …, 168} ∵ No any codeword cj from frequently-used part satisfies d(c2, cj) THL = 135 ∴ Add c2 into frequently-used part and delete c2 from infrequently-used part cn+1={220, 212, 182, 179, 240, …, 246}

Codebook Refining (Cont.) Frequently-used part C3’={214, 85, 185, 191, 164, …, 120} C3’={214, 85, 185, 191, 164, …, 120} c2={85, 170, 166, 169, 86, …, 168} c2={85, 170, 166, 169, 86, …, 168} cn+1={220, 212, 182, 179, 240, …, 246} : : Generated codebook Infrequently-used part cn+1={220, 212, 182, 179, 240, …, 246} ∵ No any codeword cj from frequently-used part satisfies d(cn+1, cj) THL = 135 ∴ Add cn+1 into frequently-used part and delete cn+1 from infrequently-used part Empty

Image Compression Generated codebook : Index table Image 1 w h 1 C0={214, 85, 185, 191, 164, …, 120} c1={85, 170, 166, 169, 86, …, 168} : cn={220, 212, 182, 179, 240, …, 246} Index table Image

Experimental Results (a) Alan (b) Goldhill (c) Lena (d) Tiffany

Experimental Results (Cont.) (a) Codebook size comparison

Experimental Results (Cont.) (b) PSNR comparison with different THs

Experimental Results (Cont.) Methods Images Proposed method (TH = 25, THH =30, THL = 115) VQ PSNR CR number of codewords Alan 31.82 0.0547 77 31.75 0.0625 256 Boat 28.68 229 28.85 Goldhill 28.96 180 29.55 Jet 29.47 178 29.81 Lena 30.27 170 31.15 Pepper 30.56 177 30.94 Tiffany 30.87 107 30.92 Toys 30.53 233 29.10 Zelda 33.320 136 34.29 Average 30.50 0.0608 165.2222 30.71

Conclusions The proposed method significantly reduces the size of codebook and maintains a good image quality of the compressed images.