Segmentation and Shape Analysis for Tracking in TV Video

摘要

Techniques of automatically segmenting tactical targets from their background in image data and then classifying the extracted shapes have been explored. The segmentation method uses grey level, edge content, and texture. Primary research emphasis has been on improving the method of shape analysis using Fourier descriptors. A new method of classification in Fourier descriptor space has been implemented which allows quicker and more accurate library searches while requiring less storage. Also the concept of partial shape description has been further developed during this research period. (Author)