A Comparison of Two Low Bit Rate Image Coders

This paper presents an experimental comparison between a VQ based coder and a classical transform coder. In the VQ coder each 4x4 image block is vector quantized using Self-Organizing Feature Maps (Kohonen Maps) which can be interpreted as structured codebooks. The codebook structure allows the use of fast search procedures which are described and experimentally evaluated. The transform coder uses the DCT to decorrelate the image values in each 8x8 block. The DCT coefficients are then encoded using a bank of Max-Lloyd nonlinear quantizers followed by an arithmetic coder. Experimental tests are performed at bit rates between 0.5 and 0.75 bit/pixels. It is concluded that similar objective and subjective performances were achieved by both coders with the transform coder performing slightly better.

[1]  Joel Max,et al.  Quantizing for minimum distortion , 1960, IRE Trans. Inf. Theory.

[2]  Nasser M. Nasrabadi,et al.  Image coding using vector quantization: a review , 1988, IEEE Trans. Commun..

[3]  Teuvo Kohonen,et al.  The self-organizing map , 1990 .

[4]  R. Gray,et al.  Vector quantization , 1984, IEEE ASSP Magazine.

[5]  Scott E. Budge,et al.  Compression of color digital images using vector quantization in product codes , 1985, ICASSP '85. IEEE International Conference on Acoustics, Speech, and Signal Processing.

[6]  Glen G. Langdon,et al.  An Overview of the Basic Principles of the Q-Coder Adaptive Binary Arithmetic Coder , 1988, IBM J. Res. Dev..

[7]  Ronald E. Crochiere,et al.  Frequency domain coding of speech , 1979 .

[8]  R. M. Mersereau,et al.  Structural image codebooks and the self-organizing feature map algorithm , 1990, International Conference on Acoustics, Speech, and Signal Processing.