ALGORITHMS FOR GENERATING IMAGE SEGMENTATION MAPS FROM SUPERPIXELS BASED ON A COMBINED INFORMATION QUALITY MEASURE
D. M. Murashov
a:2:{s:4:"TYPE";s:4:"TEXT";s:4:"TEXT";s:105:"Federal Research Center “Computer Science and Control” of the Russian Academy of Sciences, Moscow, Russia";}
Keywords: image segmentation, information redundancy, variation of information, combined quality measure, segment growing, superpixel merging
Abstract
This work proposes new algorithms for combining superpixels into segments using a “greedy” strategy and a combined quality measure that includes two components: a measure of information redundancy and variation of information. The “greedy” strategy was previously used by the author to speed up image segmentation from the condition of minimum information redundancy, and a two-component information quality measure was used in the problem of combining different segmentation maps. The joint use of the “greedy” strategy and the combined quality measure in new algorithms is aimed at accelerating the generation of segmentation maps and improving their quality through a compromise between the requirements of minimizing the number of informationally important segments and minimizing the information difference between the original images and the generated partitions. A computational experiment on test images shows that the proposed algorithms can speed up the segmentation process compared to the method based on minimizing information redundancy previously used by the author and improve the information characteristics of the resulting segmentation maps.
|
