Texture feature extraction in the wavelet compressed domain

The necessity of compression for storage coupled with the requirement of efficiently processing the compressed data has led to a great need for compressed-domain methods in which data can be processed without decompression. One procedure frequently applied to image data is the calculation of texture features for representing image content. The extraction of texture features from compressed images can be cumbersome using traditional decompress-process methods since decoding and storage of the data is necessary. This dissertation proposes a method for calculating wavelet energy texture features directly from a wavelet-compressed symbol strewn. This new method is based on a wavelet-based coder called the Embedded Zerotree Wavelet Coder (EZW) [Shapiro93], which supports high-quality compression and progressive transmission. The EZW is selected as the framework for developing algorithms because of these properties. The proposed method requires little decompression (only arithmetic decoding is necessary) and results in a technique that is fast and requires less memory than traditional approaches. These savings are achieved by eliminating the need to reconstruct and store the original image and through the simplification of the operations performed. The work in this dissertation addresses the current state of compressed-domain technology and describes how this research contributes to the field's progress. The developed algorithms have been implemented at various compression ratios, and for each case the classification results are nearly identical to those obtained with the traditional method.