Extracting edge information is the essential process in most pictorial pattern recognition problems, since subsequent shape and size measurements can then be taken. The existing techniques for extracting the edge information include thresholding methods, contour following methods, and combinations of both. The shortcomings of the existing boundary detection techniques are that they are either unreliable for detecting boundaries in a noisy background or too specific to cover a broad class of problems.

In this paper, a computer algorithm for boundary detection will be presented. First, a criterion function is set based on the specific applications at hand to evaluate the performance or cost of detecting the boundary at the local and global levels. Then the set of boundary points is selected as the ones which minimize this criterion function. This algorithm has been tested to locate the heart boundary on PA chest x-ray pictures. It is found that this algorithm will successfully extract the heart boundary even in very noisy backgrounds.