摘要：The aim of the work is to develop an algorithm forextracting local extrema of images with low computationalcomplexity and high accuracy. The known algorithms for blocksearch for local extrema have low computational complexity,but only strict maxima and minima are distinguished withouterrors. The morphological search gives accurate results, inwhich the extreme areas are formed by non-strict extrema,however, it has high computational complexity. This paperproposes a block-segment search algorithm for local extrema ofimages based on space-oriented masks. The essence of thealgorithm is to search for single-pixel local extrema andregions of uniform brightness, comparing the values of theirboundary pixels with the values of the corresponding pixels ofadjacent regions: the region is a local maximum (minimum) ifthe values of all its boundary pixels are larger (smaller) orequal to the values of all adjacent pixels. The developedalgorithm, as well as the morphological search algorithm, allowto detect all single-pixel local extrema, as well as extreme areas,which exceeds the block search algorithms. At the same time,the developed algorithm in comparison with the morphologicalsearch algorithm requires much less time and RAM.
关键词：local extrema of images; strict and non-strict extrema; block-segment search for local extrema; image segmentation; space-oriented masks; region growing
