Study on the Choice of Wavelet Filters for Image Compression using Neural and k-Nearest Neighbor Classifiers

In wavelet based image compression techniques, the selection of filters influences the compression performance. The filter suitable for one image needs not be the best for another. Whether one can relate one or more image properties with the filter selection is of great importance. The objective of this paper is to examine the influence of image properties on the selection of suitable wavelet filters. A set of spatial, spectral as well as statistical features of images are considered. The images are compressed by applying different families of wavelets. For a given image, a wavelet filter giving the best performance is decided based on the compression ratio and quality of the reconstructed image. The choice of filter for each of 183 test images was found, and this information was given to different classifiers like neural, nearest neighbor, discriminant and classification tree to predict the best filter for image compression based on image features. The study reveals that the parameters used in this investigation are not sufficient for the selection of appropriate wavelet filters.