The Domain of Orthogonal Transforms and the Understanding of Image Features

In this study, we explore the domain of orthogonal transforms, in order to bring an understanding on the characterization of image features, with emphasis placed on the Karhunen-Loeve (K-L) transform for its optimal energy packing properties. The study's wnmbution is in establishing a thorough research base that relates the eigensystem and transform domain properties of the K-L transform to two-dimensional image features. Other transformations such as Haar, Hadamard, Walsh, Fourier, and the discrete cosine transform ( D O are presented as a basis for performance evaluation in the energy packing sense. The main thrust of this study, given the transform domain, is to determine mechanisms and develop algorithms relevant to the understanding and discrimination of visual features invariant across translation, orientation, gray-level polarity reversal, and size. The computational requirements are addressed.