Application Research On Digital Image Segmentation With Improved Genetic Algorithm

Image segmentation is the basic technology in image processing and computer pre-vision, it's an important component in most image analysis and vision system, the efficiency of the image segmentation directly affect the follow-up work of the image processing. The threshold method can implement simply and quickly complete the image segmentation by some threshold selection criteria. It's an efficient image segmentation. But for different image, the different segmentation criteria has different segmentation effect, furthermore when the particularity segmentation is needed for the image information, one threshold is obviously not enough for the requirement, while the actual effect of the multi-threshold segmentation using the exhaustive search method is bad, it can't meet the requirement for the actual application. It must choose a fast and exact optimization method combined with the threshold method for the image segmentation.Genetic algorithm (GA) is an intelligent algorithm to solve the complex optimization problems, which can be fully competent for the area of the image segmentation. GA can quickly achieve the optimization and has a good robustness, which can be the same with the various threshold criteria. However, the randomicity and easily early convergence restrict its application. It needs to do some improvement to improve the performance of the convergence.In this paper, it firstly analyzes some general method of image segmentation, focusing on comparative analysis of the threshold methods. Based on the previous study, some improvement in the cross-entropy segmentation method is implemented and a new cross-entropy segmentation method is proposed. At the same time, combined with some general change in the GA, an improved genetic algorithm is proposed. It's using the non-linear adaptive crossover probability to improve the crossover operator, adjusting the mutation probability by the convergence situation to improve the mutation operator, besides it joins the best individual preservation, the population recombination strategy by replacing the worst individual with the random individual. After these measures, it improves the performance of the algorithm. A test has been done by using some common algorithm testing functions. Through the result compare and analysis with the SGA, AGA, it proves that the improved GA has been greatly improved in the convergence speed and accuracy. In the end, the improved GA is combined with the Otsu method, the maximum entropy method and the proposed cross-entropy method in the image segmentation. By increasing the number of the threshold, compared with the conventional exhaustive search method, the results show that the improved GA can do well in the multi-threshold image segmentation, moreover the more numbers of the threshold is increased, the more preponderant the algorithm reflects. Not only is the actual effect good, but also the accuracy is very high, which can be fully used in the actual project. Meanwhile, the segmented image also verifys the validity of the cross-entropy method proposed in this paper.