Parallel dynamic programming algorithms for image recognition based on its contour

Abstract

The dynamic programming is an important procedure for speech understanding and image recognition. In this paper, the application of dynamic programming to recognize the image basing on its contour is recalled, then the parallelism of this process is analyzed and exploited. Some parallel dynamic programming algorithms for evaluating the smallest value of the minimum intersegment distance path in a data table are designed in detail on several kinds or the specialized parallel computer, such as 2D Systolic Array, ID Systolic Array and an associative computer of the SIMD type. Their computation complexities are respectively estimated as O(k), O(k) and O(kw), while the cost O(mn) is obtained on serial computer, where k=max(2n,m), m is the number of segments of a contour to be recognized, n gives the number of segments of the sample contour and w is the bit-width of the used associative fields.