Context-Based Entropy Coding of DCT Coefficients for Image Compression

• Main Authors: Yu-Mao Hung, 洪毓懋
• Other Authors: Shin-Hsu Chang
• Format: Others
• Language: zh-TW
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/32184015815039621210

碩士 === 大葉大學 === 資訊工程學系碩士班 === 95 === This paper proposes Context-based entropy coding of DCT coefficients for image compression. The main compression step is: （1）Divide the input image into the 8×8 block, use Discrete Cosine Transform（DCT） for every block; （2）Find out suitable quantization coefficient, quantize DCT coefficient; （3）Use the Direct Current coefficient（DC） to predict that Alternating Current（AC） coefficient, get AC error coefficient; （4）Use the differential pulse code modulation（DPCM） to DC coefficient, get DC error coefficient; （5）Context coding is used in DCT coefficient. Models of Context Coding: （1） Zero Coding, （2） Refine Coding, （3） Run-length Coding. When enter compression step, quantization coefficient will be recomputed. The performance which doesn’t need extra quantization table is better than the performance which needs to use fixed quantization table. When using arithmetic coding, consider the relation of coefficient can improve the performance of compression. According to the characteristic of Discrete Cosine Transform, using DC coefficient to predict AC coefficient can improve the performance of compression. Because coefficient transform is one kind of block-based coefficient transform. When image would be compressed overly, the restructure image will cause distinct Blocking Effect. Post-processing can improve the performance of compression.