3D balance quantization encoding based on variable size matrix segmentation

• Published in: 2010 International Conference on Information and Automation pp. 2325 - 2328
• Main Authors: Aijun Sang, Hexin Chen, Gabbouj, Moncef
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.06.2010
• Subjects: 3D Balance Quantization; Discrete cosine transforms; Discrete transforms; Encoding; Entropy coding; Humans; Image coding; image encoding; Image segmentation; Quantization; three dimensional matrix; Transform coding; variable size segment; Visual system
• ISBN: 1424457017; 9781424457014
• DOI: 10.1109/ICINFA.2010.5512427

Variable size matrix segmentation and variable size transform are used in the colour image compression methods based on three dimensional matrix widely discrete cosine transform (3DMWDCT). However the coefficients have different statistical properties since different areas of an image and difference size blocks have difference energy centralization. 3D balance quantization based human visual system was proposed in this paper. After dividing the original colour image into variable sized sub-three dimensional matrices according to the activity characteristics of the image, the variable sized 3DMWDCT is applied to the corresponding sized sub three dimensional matrices. DCT coefficients were 3D balance quantized through non-linear scalar quantization and entropy coding. Experimental results show that the compression efficiency of the proposed algorithm is better than that of JPEG. Compared with JPEG, the proposed algorithm improves PSNR at most 2dB, and the quality of the reconstructed image is improved in subject evaluation too.