A Scheme for Image Classification and Adaptive Mother Wavelet Selection

Abstract

Wavelet based image coders are gaining popularity. The performances of such image coders are highly depend up on the selection of an appropriate wavelet for a particular image. There are many suggestions to classify images in to classes. In this paper, we are introducing classification of the images based on their statistical properties. Once the image is assigned to a particular class based on their statistical properties, then an appropriate wavelet is to be selected for the compression. A technique called adaptive mother wavelet selection scheme is proposed. The approach is based on the classification of images into different groups and associating each of the groups with the best wavelet that provides best quality images for a given amount of compression. Performance quality measures like PQS, bit rate, PSNR are used for judging the best wavelet for a class of images. Image properties like mean, standard deviation, and skewness are used for classifying images into groups such as texture, human, and aerial. Implementation results on test images demonstrate the superiority of the proposed technique.