ABSTRACTS OF ARTICLES OF THE JOURNAL "INFORMATION TECHNOLOGIES".
No. 4. Vol. 26. 2020

DOI: 10.17587/it.26.252-256

G. K. Bukalov, Professor, e-mail: gk.bukalov44@yandex.ru, A. O. Burygin, PhD student, e-mail: g.t.m.p@yandex.ru, I. G. Panin, Professor, e-mail: igpanin@list.ru, Kostroma State University, Kostroma, 156005, Russian Federation

Application of Community Building Methods for Segmentation of Textile Slings Images

There is problem of segmentation of textile slings by graph methods of community detection is considered. Image is initially segmented by the Meanshift algorithm, followed by the construction of a Weighted Region Adjacency Graph (WRAG), the vertices of which represent the regions obtained after the initial segmentation. The quality of the graph partitioning into subgraphs is determined by the Newman criterion. Edge weights are calculated based on the color and texture characteristics of the image region. Comparison of graph node weight metrics: color similarity metric defined by Mean Distance, texture property metric defined by Histogram of Oriented Gradients, and superposition of LAB and HOG image components. The FMCDRN algorithm is used to detect communities on the graph. Each community has a mask of the real object in the image. To determine the effectiveness of the proposed system, are use Probabilistic Rand Index (PRI), Variation of Information (VOI), and F-measure. Comparison of the effect of initial segmentation by Meanshift and Superpixel algorithms. Qualitative comparison of FMCDRN, Infomap, FGMDO, Louvain methods for highlighting communities on a graph. A computational experiment aimed at studying the effectiveness of the proposed method is carried out. Comparison of the proposed algorithm with modern image segmentation frameworks WiseCode, LC, EDISON.
Keywords: community allocation on graphs, image segmentation, Region Adjacency Graph, histogram of oriented gradients (HOG), Newman criterion

To the contents