A Hadamard Transform Fuzzy Segmentation And Classification Technique For Image Compression

Abstract

A new technique for image data compression based on fuzzy segmentation, classification and properties of sequency ordered Hadamard Transform (HT) is presented. Using a pyramidal data structure, an image is segmented into blocks of variable sizes. The size of each block is determined using a fuzzy procedure depending on the amount of information contained. Regions with low details are divided into blocks of variable size, while regions containing more information are successively segmented into smallest size of 4*4 blocks. Only a small number of more energetic components of HT are used for the coding of the uniform blocks, while a fuzzy edge oriented classifier is designed for the coding of the high-detail blocks. The classifier employs several prototype edge patterns in HT blocks and the fuzzy techniques for recognition of the direction of edges. The low-detail regions are coded with very low bit rate on the expense of small reduction in the visual quality of images, while the coding process for high-detail regions results in an acceptable image quality. Decoded images of high quality are obtained for encoding rates of 1 bit per pixel.