Design and Performance Analysis of Linear Phase Para-UnitaryMBand Filter Banks for Image Coding

This paper presents the optimal design and performance analysis of the linear phase paraunitary (LPPU)Mband filter banks in the frame of vector quantized image coding. First, by maximizing the unified coding gain, which is a function of intra-band correlation, as well as inter-band energy compaction, the LPPUMband filter banks for the general factorization form are designed. Then, the image coding performances of the LPPUMband filter banks, such as the inter-band energy compaction, the intra-band correlation, and the average entropy are discussed. It is shown asymptotically that, as the filter length increases, the unified coding gain for the LPPUMband filter bank improves and the unified coding gain of the LPPU 4 band filter bank approaches very closely that of the LPPU 8 band filter bank. This observation is also verified by extensive computer simulation on the real images. In addition, the benefit of the directMband decomposition, based on the LPPUMband filter bank, over the tree structure decomposition is discussed, by showing the comparable coding performance and the reduced computational complexity.