Development of shape analysis algorithms

This report summarizes the results of a 4 year research effort. The primary research topic addressed is automatic shape recognition of silhouettes or contours. Methods studied are Fourier descriptors, Walsh descriptors, and 2-D moments. Also the recognition of partially correct shapes are addressed. Three methods were developed to solve such problems. These methods are: (1) polygonal fitting of a boundary followed by local matching of distances and angles; (2) Fourier descriptor representation of boundary segments followed by dynamic programming to match segments (dynamic programming is modified to allow several local optimums instead of one global optimum; and (3) recognition of time-axis scaling and matches between two partially matching waveforms using Mellin Transform techniques. Other research topics addressed by this summary are: (1) registration and segmentation of moving objects; (2) measurement and recognition using 3-D range data; (3) image compression using motion detection; and (4) precision measurements in digital images. The research summarized in the report resulted in 5 Ph.D. Theses, 5 M.S.E.E. theses, and over 20 additional publications.