Region based subband image coding scheme

The statistical properties of image subbands are studied for various subband decompositions. The authors observe that the amount of energy in image subbands increases towards the lower frequency subbands and, more importantly, towards the image edges, simultaneously. They also find that the directionality of energy distribution is highly dependent on the orientation of the edges. Based on the energy distribution, they extract arbitrarily shaped regions in each subband and employ the entropy-constrained quantizers using the generalized Gaussian distribution for modeling the image subbands. They also address the problem of determining an optimal subband decomposition among all the possible decompositions. The resulting feature based subband coding scheme yields at least 1-dB better PSNR than previously proposed schemes for the "Lena" image.<>