Approaches to two-dimensional object recognition
Cho-Hauk Teh · 1988
This thesis presents some approaches to two-dimensional object recognition. The objectives are to handle object translation, rotation, scale change, and occlusion. First, various types of moments, including geometric, Legendre, Zernike, rotational, complex, and pseudo-Zernike moments, are evaluated and compared. Three fundamental issues are investigated both analytically and experimentally. They are (1) sensitivity to image noise, (2) aspects of information redundancy, and (3) capability for image representation. Next, a scale independent method is developed to locate dominant points (points of high curvature) on a boundary curve, which are joined to form a polygon to represent the object boundary. The method requires no input parameter and remains reliable even when features of multiple sizes are present on the boundary curve. It first determines the region of support for each point on the boundary based on its local properties, then computes measures of relative significance (e.g., curvature) of each point, and finally locate dominant points by a process of nonmaximum suppression. This leads to an important observation that the performance of dominant points detection depends not only on the accuracy of the measure of significance, but mainly precise determination of the region of support. This solves the fundamental problem of scale factor selection encountered in various dominant point detection algorithms. Finally, a parallel local feature aggregation method is developed for recognizing two-dimensional objects based on geometric models (e.g., CAD models). The method can handle object translation, rotation, scaling, and occlusion. Two types of local features, the L structures and the U structures, are extracted from the input gray-scale or binary image and matched with those of the model. Each match hypothesizes the object location in the input image, and a similarity measure is computed to indicate the probability of the match. A large accumulation of the similarity measures at a hypothesized location indicates probable existence of the object in the input image.