A Kohonen-based structured codebook design for image compression

The Kohonen neural network (KNN), creates a vector quantizer by adjusting the weights from input nodes to output nodes. Using this model, we obtain a codebook and also the neighborhood relations between the codewords of this codebook. Based on these neighborhood relations, we can create a structured codebook to improve the search time and/or bit rates. However, there is an intrinsic problem of applying the KNN directly. When the codebook size increasing, the coding performance is not good uniformly. To overcome this bad situation, we propose a modified version of the KNN, called the hierarchical KNN (HKNN). Using this method, the image "lenna" (of size 512/spl times/512) can be coded at 0.5 bpp with PSNR 32.197 dB. Besides, we also present an adaptive VQ scheme, adaptive HKNN, for image sequence coding. According to our experimental results, the improvement of adaptive HKNN can be up to 2.5 dB with 0.16 bpp transmission overhead for the image sequence "claire".<>