Multi-level image thresholding based on local variance and particle swarm optimization

Multi-level thresholding is a basic pre-processing technique in computer vision and pattern recognition tasks. Optimal threshold values classify pixel values into multiple categories. The proposed method performs multi-level grayscale image thresholding based on local variance of pixels and Particle Swarm Optimization (PSO). Local variance contains foreground and background variances. The main idea of the proposed technique is combination between Sobel ability, foreground and background variances in order to provide higher visual perception. This novelty causes better exploration of search space in order to find appropriate threshold values for region uniformity. Experimental results show the superiority of this approach in comparison with other thresholding approaches.