Research On Image Threshold Segmentation Algorithm Based On Maximum Entropy And Genetic Algorithm

Image segmentation has always been a classical and important research subject in the field of image processing, now it is becoming the key technology of computer vision and pattern recognition. The purpose of image segmentation is to divide the image into a plurality of regions with same characteristics and extract the part we are interested in according to the specific criteria. The results directly influence the implementation of subsequent image analysis and image understanding. There are many different types of segmentation methods at present. According to the types,they can be divided into image segmentation based on region, image segmentation based on edge detection, image segmentation based on threshold and image segmentation based on specific theory.This paper only introduces image segmentation method based on threshold carefully as we have different emphasis. The realization principle of threshold method is to find out the contrast between the target and the background in the image by the existing histogram of the image and determine the threshold to separate the target and the background. The experimental results show that the threshold method is a simple, fast and efficient method of segmentation.The classical one-dimensional Otsu method and the maximum entropy method are good at with segmenting gray level histogram markedly bimodal or target and background is in marked contrast, but if the image is affected by noise and the histogram becomes uniform, the results of segmentation is not ideal.In order to obtain better threshold in image segmentation, this paper expounds a segment method combining maximum entropy image with genetic algorithm. Genetic algorithm is derived from the theory of Darwin's "survival of the fittest, the survival of the fittest" in evolutionary genetic theory. In the nature of the process of biological evolution, biological populations of individuals through genetic and mutation to adapt to changes in the external environment, inferior individuals can't adapt to change gradually be eliminated, the excellent individuals can be preserved. Genetic algorithm is a global optimization algorithm for optimizing the maximum entropy threshold segmentation. The implicit parallelism can speed up the search speed of the threshold, but the experiments found that the threshold fluctuations larger and sometimes convergence rate is very slow or no convergence. One dimension maximum entropy does not consider the neighborhood pixel gray level, threshold is not very accurate.In order to solve these two problems, one-dimensional maximum entropy is extended to two dimensions. The two-dimensional histogram of gray level-neighborhood mean gray value is used to calculate the entropy. As the algorithm considers the spatial information of pixels, it can get more accurate threshold. But it needs to traverse all of the data to get the optimal threshold segmentation which leads to large amount of calculation and high time complexity. This does not meet the requirement of real-time processing.Then the improved genetic algorithm for global optimization of two-dimensional maximum entropy image segmentation is introduced. By improving the genetic operators, the algorithm can adjust to control parameters of population evolution according to the population individual adaptive value and population characteristics. It can not only maintain the diversity of the population, but also accelerate the speed of convergence. This method overcomes the problem that the convergence of traditional genetic algorithm is poor or not convergent. Comparing with one-dimensional maximum entropy method, maximum between class variance method and one-dimensional maximum entropy combined with traditional genetic algorithm and two-dimensional maximum entropy, for image 196073, the contrast of region and regional consistency of 2D maximum entropy combined with improved genetic algorithm is the highest, respectively, of 0.2525 and 0.7608. At the same time, threshold searching time compared with the 2D maximum entropy method is also greatly reduced, compared with the one-dimensional maximum between class variance and the maximum entropy respectively reduction of 6.96%?and 2.81%. The result is similar for image 15033,2-D maximum entropy combined with improved genetic algorithm has the highest contrast of region and regional internal consistency and the threshold searching time is reduced 5.37%?and 0.17%?respectively compared with the one-dimensional maximum between class variance and the maximum entropy. What's more, the algorithm is more stable and has more convergent speed compared with 1-D maximum entropy, which shows that the method has practical value.