Recognition of kidney glomerulus by dynamic programming matchingmethod

Dynamic programming was applied to locate the glomeruli in microscopic images of kidney tissue section. The glomeruli were modeled by a polygon whose sides could be varied within a given range of lengths. The objects were located by determining the best match of the model according to a so-called optimum criterion in which all possible shapes were evaluated at all possible positions in the input image. The best model was selected according to the maximum average gray level. To increase the probability of obtaining a closed contour, a distance criterion was added and the maximum gray-level requirement was relaxed somewhat. The optimum criterion was modified to include a directionality constraint in which the difference in angle between model segments and the edge values in the image was minimized, thereby increasing the performance of the method. A hierarchical multiresolution strategy was used to reduce calculation time. The cyclical property of a contour is also taken into account