An algorithm of image segmentation based on community detection in graphs

This article suggests an algorithm of image segmentation based on the community detection in graphs. The image is represented as a non-oriented weighted graph on which the community detection is carried out. Each pixel of the image is associated with a graph vertex. Only adjacent pixels are connected by edges. The weight of the edge is defined by subtracting the intensities of three color components of pixels. A Newman modularity function is used to check the quality of the graph partition into sub-graphs. It is suggested that a greedy algorithm should be applied to solving the image segmentation problem. Each community corresponds to a segment in the image. A computer experiment was carried out. The influence of the algorithm parameter to the segmentation results was revealed. The proposed algorithm was shown to be insensitive to random impulse noise.