Unconstrained-endpoint dynamic space-warping algorithm with experiments in binary English character images

Abstract

The unconstrained-endpoint dynamic space-warping algorithm (UEDSWA) is presented for two-dimensional template matching. The UEDSWA computes three kinds of distances: pixel-distance, line-distance and area-distance. Pixel-distance is the distance between two pixels and is computed by using a pixel-distance measuring scheme. Line- and area-distances are the distances between two line segments and rectangular areas, respectively. Both distances are computed using dynamic programming techniques based on the principle of compression and expansion. The UEDSWA computes a minimum distance from two image templates. By using binary English character images, experimental results show that the minimum distances computed by the UEDSWA have great potential for measuring the similarities between binary images.