Nonseparable orthogonal linear phase perfect reconstruction filter banks and their application to image compression

We have designed a 4-channel 10×10 nonseparable orthogonal linear phase perfect reconstruction filter bank (PRFB), hereafter NSEP, and examined its performance in image compression applications. It is shown that, in addition to the orthogonality property, NSEP offers an excellent compression performance. Because of their limited spatial support, the nonseparable filters can be implemented at a reasonable computational cost. The orthogonality property also allows the use of image compression algorithms which exploit orthogonality for bit allocation purposes. We demonstrate that the NSEP filters have superior spatial localization properties when compared to filters arising in a 2-D separable PRFB generated from a l-D Daubechies FB of length 10. This superior spatial localization results in less visible ringing distortion in compressed images. In addition, the linear phase property of NSEP distributes the ripples located at image edges evenly on both sides of each edge, thus reducing further their visibility. Because of these advantages, NSEP is a viable and attractive candidate for high quality, low bit rate image compression applications