Adaptive dichotomous image segmentation toolkit
Mikhail Vyacheslavovich Kharinov · Pattern Recognition and Image Analysis · 2012
The article deals with the adaptive hierarchical segmentation of the digital image divided into segments of the calculated form. The dichotomous segmentation is examined, in which case the vast majority of segments are divided into two nested segments. A method is described for the rapid construction of a dichotomous hierarchy by a given criterion for the proximity of segments by the iterative merging of adjacent segments of an initial image decomposition. The numerical characteristic of the regularity of the dichotomous hierarchy of segments is introduced. Variants are constructed of the hierarchical approximation of the image by nested partitions (levels of the hierarchy) formed from the segments with repetitions. Three basic types of transformations are determined on the set of hierarchical decompositions. In order to optimize the approximation of visible objects by image segments, 22 algorithms are studied of its partition to successively increase the number of segments. Depending on the number of segments, the required standard deviation is estimated. A comparison with similar solutions is given.