Membrane Computing-Inspired Clustering Algorithm And Its Application In Image Segmentation

Clustering can be defined as the partitioning of a given set of m data points into c classes such that data in the same group are as similar as possible and data from different groups are as dissimilar as possible, which is an unsupervised learning process. Differential Evolution algorithm is a random searching algorithm.Image segmentation can be seen as the process of dividing an image into several pixel regions so that in the same pixel region is as similar as possible and each pixel region is as dissimilar as possible. The purpose of image segmentation is to detect the foreground and the background. Clustering algorithm is widely used in image segmentation.Membrane computing is also known as P system, which is inspired from the structure and functioning of living cells as well as interaction of living cells in tissues and organs. In the past years, a large number of variants of P systems have been considered and applied to solve a variety of real-world problems.This paper focus on discussing a kind of clustering algorithm inspired by membrane computing and the application to image segmentation, the main innovations are shown as follows:(1) DE-MC algorithm. The designed cell-like P system has a two-layer membrane structure, three differential operators(mutation, crossover and selection) in differential mechanism are introduced to achieve the evolution of objects and the communication mechanism of the cell-like P system is applied to share the objects between different membranes. The DE-MC algorithm is evaluated on the artificial and the real-life data set and is further compared with classical k-means algorithm, DE-based clustering algorithm and GA-based clustering algorithm respectively.(2) An image segmentation algorithm based on P systems. Under the cell-like P system of three-layer nested structure, it is to finish the image segmentation. The algorithm is evaluated on several grayscale images and is further compared with a variable length genetic algorithm-based fuzzy clustering technique and an adaptive differential evolution-based fuzzy clustering algorithm respectively.