The Boundary Processing of Wavelet Based Image Compression

When an image, which is decomposed by bi\|orthogonal wavelet bases, is reconstructed, some information will be lost at the four edges of the image. At the same time, artificial discontinuities will be introduced. We use a method called symmetric extension to solve the problem. We only consider the case of the two\|band filter banks, and the results can be applied to M\|band filter banks. There are only two types of symmetric extension in analysis phrase, namely the whole\|sample symmetry (WS), the half\|sample symmetry (HS), while there are four types of symmetric extension in synthesis phrase, namely the WS, HS, the whole\|sample anti\|symmetry (WA), and the half\|sample anti\|symmetry (HA) respectively. We can select the exact type according to the image length and the filter length, and we will show how to do these. The image can be perfectly reconstructed without any edge effects in this way. Finally, simulation results are reported.