Polygonal Approximation Using A Multiresolution Method And A Context-Free Grammar

An new method to obtain polygonal approximation for object contours is presented. The method consists of coding the contour with the Angle Freeman chain code (AF8), obtaining strings of eight symbols, and looking for patterns of substrings that represent slope changes along the contour. Our strategy for detecting dominant points is to look for the ends of discrete straight lines through patterns of AF8 symbols, which can be produced by a context-free grammar. With a multiresolution method, we present the polygonal approximation for noisy contours. A set of N dominant points is obtained, the integral square error (ISE) is calculated and, finally, based on lost pixels (LP) in decoding process a new error criterion, that we call lost ratio (LR), is proposed. We found that our method produces the lowest ISE, LP and LR regarding the state-of-the-art.