Divergence, householder transformations, and information preservation



Journal Title

Journal ISSN

Volume Title



One of the major problems in the field of pattern recognition is the dimension of the feature space. The cost and efficiency of a classification technique depend upon the number of features. It is essential to develop a feature compressor that will transform the features into a space of lower dimension without significant loss in information. The major concentration, in this thesis, will be the development of a compression transformation matrix B. Householder Transformations will be utilized to construct a matrix B and the divergence criterion will be used to measure the effectiveness of transformation's ability to preserve information. Several preliminary theorems and definitions will provide a foundation for feature compression. Agricultural multispectral data will be used to demonstrate the utility of the procedure for various dimensions of the transformed space.