Intelligent program for automatic image recognition based on compact object-fitting hierarchical image representation in terms of dynamic irregular branchy trees

Adaptive dichotomous image segmentation toolkit

“…Since the regularity index θ is calculated without taking into account repetitions of segments, it matches the indices θ C and θ E , which are calculated for the par titioning hierarchy in compact and expanded views (7) In addition to the fact that isotonic transformations C and E do not affect the regularity index, they also satisfy the symmetric commutation relations…”

“…Since, in the dichotomous hierarchy, individual segments can be divided into any number of nested segments, their generation, storage, and conversion are conducted in terms of the overall structure of the data of dynamic trees [7,8], which supports any hierarchy of nested partitions or image approxima tions by representations with averaged values of brightness.…”

“…It is characteristic that, as the iterative computation of nested partitions of the image proceeds, the speed of each subsequent iteration increases and the total time of constructing the hierarchy grows approximately as the cube root of the number of iterations. Thus, according to our experimental data, the time of the hierarchical seg mentation generation depending on the number of The partitioning hierarchy of the image is called compact if, at different levels, it does not contain repeating composite (divisible) segments [7,8]. At the same time, only repetitions of elementary segments grouped on levels with a large number of segments are permitted.…”

“…1 We began the research on the adaptive hierarchical segmentation somewhat later and, to compensate for the lack of memory and computer speed of that time, we developed a special data structure, so called "dynamic trees" [7,8], which is still relevant to the implementation. Currently, because we used results from [3][4][5][6], we managed to significantly improve the quality of the adaptive hierarchical segmentation obtained by fast algorithms [8].…”

“…In contrast to the existing stereotype [6,7], in this paper, we consider the "almost" regular adaptive image segmentation, in which segments are divided primarily into two nested segments, and the hierarchy of consecutive partitionings of the image in the tradi tional terminology is described by the trees, in most nodes of which not more than two arcs are merged. The consideration of the transformation of partition ing hierarchies in one another, carried out in terms of dynamic trees, is also a feature of the proposed solu tion.…”