Abstract: In this paper, the decomposition of the image into 10 subbands is realized. The image is decomposed rowwise first, and columnwise afterwards, giving four subbands. Then, the lowest subband is two times recursively decomposed rowwise and columnwise giving additional seven subbands. The coding of subbands is realized as follows: the lowest subband is scalar quantized into 256 levels, and the other subbands are vector quantized using 2×2 or 4×4 blocks. For each subband, which correspond to the some specific orientation of the image, the code table is constructed. The length of code table is 16, for the higher subbands, and up to 256, for lower subband. These code tables are optimized using training set of many different images. The test image "Lena" is not contained in the training set, hence, the code tables do not increase the entropy of the coded test image. The experimental results show that the subjective quality of the image is very good, at the first order entropy of about 0.33 bpp and PSNR of 29 dB.
Key words: Image coding, vector quantization, image decomposition, code table, Lena, filter, quadrature mirror filter, PSNR.