Abstract
One of the fundamental problems of shape discrimination by machine vision is the object matching based on the boundary information. Subsequently, precise edge detection has been the main issue in this field. New techniques to improve the precision of edge detection and object matching are introduced and a series of computer experiments and the successful results are described in this dissertation. (1) A concept of edge point instead of edge pixel was introduced for a more precise and smooth expression of boundary silhouette. A dual edge phenomenon has been shown to be accompanied with the edge point expression of the zero crossing of the second directional derivative on facet fit. This phenomenon was treated such that a thin edged boundary which could be followed without ambiguity was yielded. (2) A scheme for edge boundary following was developed. The pixel space of edge boundary was scanned only one time producing a set of 'lists'. A list is defined by a set of pixels that are contiguous in a row or column. This lists were linked giving a 'coordinate string' for each contour. A coordinate string is a series of edge pixels or edge points ordered in a sequence contour following. (3) Three methods of ψ - s curve generation were devised. We applied them to produce three types of contour from a coordinate string. And then, those were compared in terms of length precision, smoothness of curve, and detailness of boundary expression. It was shown that the best ψ - s curve in terms of the three parameters could be obtained from the coordinate string of "AI thin edge boundary". The AI thin edge boundary was named for the edge boundary points that were obtained by treatment of dual edge phenomenon. (4) Some properties of ψ - s curves were introduced with proofs. (5) A method for the detection of common boundaries between adjacent puzzle pieces using the properties of ψ - s curve were introduced. A method of corner point detection using Ψ - s curve is also introduced. The corner points detected by this method were used as a subsidiary information for the common boundary detection. (6) A ring cluster was defined for the topological relations of the puzzle pieces. And this was constructed from the logical and physical properties of the ring cluster in relation with common boundaries. The logical and physical properties were introduced. The assembly of a set of scattered regional pieces was performed on the basis of the ring cluster.
Kim, Hee Sung (1987). Shape discrimination and assembly of scattered regional pieces by Psi-S curve. Texas A&M University. Texas A&M University. Libraries. Available electronically from
https : / /hdl .handle .net /1969 .1 /DISSERTATIONS -746836.