Social influence dynamics for image segmentation: a novel pixel interaction approach

This paper introduces a novel image segmentation technique that is inspired by social influence and opinion dynamics, establishing an association between social sciences and image analysis. This methodology analogizes each pixel in an image to an individual within a population, where the intensity of the pixel reflects an individual’s opinion. By simulating social influence through iterative interactions among individual pixels, our approach emulates the interaction patterns observed in human populations. During each interaction, a pixel selects another pixel within its immediate neighborhood to compare opinions or intensity levels. If the intensities are similar, indicative of analogous opinions, we adjust their values to minimize the difference, thereby producing the formation of homogenous regions within the image. Conversely, when the intensity difference between the two pixels is significant, we manipulate the intensity of both pixels to accentuate this disparity and effectively segregate the regions within the image. After several iterations, the objects in the image tended to split according to the homogeneity of their intensities. The efficacy of the proposed technique was tested using several images and widely accepted quality metrics. The results of these experiments show that the proposed method achieves competitive performance compared to other segmentation techniques.