Highly Compressed Zernike Moments by Smoothing

A novel methodology that improves the discrimination capabilities of the compressed feature vectors, used in pattern recognition tasks, is presented in this paper. The Zernike moment signals of some patterns are extracted and are compressed by using a typical wavelet compression technique. In order to increase the compression ratio without loosing a great amount of significant information, the moment signals are smoothed appropriately before the compression. As a result, highly compressed feature vectors that include enough classification information are derived. The resulted improved features are studied for their discriminative power through appropriate experiments.