Automatic Scene Analysis Using Hierarchical Spatial Frequency Domain Correlation with Energy Normalization and Heuristic Noise Reduction

摘要

This thesis describes the development of software to perform automatic scene analysis using a correlation process. The scene analysis software is designed to locate a given target within a scene image. The operation of the scene analysis software proceeds in a hierarchical manner. The original scene undergoes analysis and sub-sections of the original scene are further evaluated as appropriate. To improve the accuracy of the correlation software, two preprocessing techniques were programmed and tested. The first technique is a technique used in earlier research efforts. This technique, edging, proved successful in improving the performance of the correlation software. The second noise-reduction technique, a heuristic technique called Queen Victoria Method (QVM), was also programed and tested. The result of tests, although nonconclusive, indicate that the QVM does not improve the results of the correlation process.