A Study Of Image Segmentation Based On The Level Set Method And Multi-objective Optimization Algorithm

As the most basic operation process in the image processing and machine vision field, image segmentation plays the vital role. Therefore, image segmentation has been widely concerned in the field of image processing. According to the gray level, color, texture and other characteristics of image has, many scholars has put forward many image segmentation methods, such as the clustering method, threshold, edge detection and so on. Most of these methods deal with the gray level of images directly, which will be affected by noise and inhomogeneous gray distribution easily. In this paper, we use a level set method which is based on region to segment image. We embed the level set function into the framework of energy function. The level set contours convergences to the optimal segmentation curve gradually in the processing of minimizing the energy function. Because traditional single objective method cannot fully consider multiple aspects of information of image segmentation,this paper adopts multi-objective method to segment image.This paper proposes two image segmentation methods. The first one uses fuzzy logic into statistical model level set algorithm. The second one applies multi-objective algorithm into the image segmentation and it be named image segmentation method based on multi-objective evolution. The contents of this paper are as follows:1. A statistical level set contours model based on the fuzzy logic is proposed. Adding fuzzy factor into the level set model can reduce the effect brought by noises and outliers, so it plays a good role at improving the robustness and accuracy of the segmentation solutions. What’s more, the introduction of the fuzzy concept and the use of regularization can protect the level set function from degenerating seriously. The use of regularization can also omit the step of reinitialion, which can improve the segment effect greatly and decrease the time complexity. The experimental results show that our method has a better result than other classic level set methods, especially at dealing with the noise points and outliers. This model has a very good effect at segmenting inhomogeneous images. This method can catch the weak boundaries effectively2. Image segmentation method based on multi-objective evolutionary algorithm is proposed and be used into SAR images segmentation. In this model, we use hard clustering function and the neighbor penalty function as two object functions. The two functions describe the image characteristics from compactness and connectedness respectively. In addition, this paper adopts the improved nondominated neighbor immune algorithm(NNIA) to optimize the two objective functions and NNIA algorithm uses crowding distance to choose the individuals from the population. So our method can avoid falling into local optimal effectively. Meanwhile, watershed transform is used into image segmentation during the preprocessing. Population initialization adopts the random coding and minimum spanning tree encoding(MST), which can improve the varieties of the population. The experimental results demonstrate that this method can overcome the local optimum and resist the noises and keep details of images effectively.