The Research On Image Segmentation Based On Improved Bp Fuzzy Neural Network

Vision is very important for human to make the perception of the world, which is the main way to get information from the world. As the basis of visual information, image objectively demonstrated to mankind everything in the world, so that people can understand the world and themselves. With the development of computer technology, it is better to help people discover knowledge by using a computer to image processing, and expand the depth and breadth of the humans’perception of the world. After decades of efforts, A relatively complete system which is called image engineering has been built, including image segmentation, image merging, image retrieval, and so on. Image segmentation is a key part of image processing, an original image is divided into several collections with it. Each collection is different from others, they do not overlap and have a strong correlation. With the collections, people can observe the microscopic world better. Essentially, image segmentation is a classification technique.During the people researching the images, they found out that image segmentation can analysis the original images deeply, that’s why the image segmentation came out and developed rapidly. For decades, thousands of methods of image segmentation have been proposed by researchers. In general, these methods can be divided into threshold segmentation, region segmentation, edge segmentation, the segmentation method with a particular theory and other kinds. The threshold segmentation, region segmentation and edge segmentation have been developed to mature. Now the research hotspot is the method with a particular theory which includes clustering analysis, fuzzy sets, artificial neural network, genetic algorithms and so on. The appearing of new methods has broadened the applied range of image segmentation, and improved the segmentation accuracy. However, there are also some problems as following:(1) The segmentation result easily gets distortion,(2) The algorithm requires more time so that it can’t be applied in large data processing.According to the above problems, this paper selects fuzzy BP neural network which is widely used in image segmentation to make the improvement. Fuzzy theory has fuzzy uncertainty, which can pick up the points from the original image. The original image can be divided into several areas which is strongly independent between others by selecting the appropriate membership functions. Every pixel is divided to classify firstly, a fuzzy value called membership is gave to every pixel which is used to measure the degree that a pixel belong to a fuzzy set. BP neural network has the commendable generalization ability, robustness and numerical approximation ability, can do the work of data processing better, parallel computing will help to reduce the time complexity. Further, it can deeply process the data from the membership function, making the membership of a pixel reflects the classification. Picking up the classification will come out a result with high fidelity.This paper selects gray level image as the object of study, picking up the feature of image as the input of BP neural network, then improving the network in several aspects, making a faster convergence. It will improve the accuracy, at the meanwhile cost less time.The contribution of this paper is as follows:(1) Introduce the study background and improved state of image segmentation, summarize the characteristic of several methods. (2) Introduce the operating principle of fuzzy theory and BP neural network. Create the fuzzy BP neural network by inputting fuzzy value, make the improvement in structure of network, transfer function, step size, training method. In the end, the improved method comes out.(3) Selects gray level image as the object of study, picking up the feature of image using the membership function. Make the simulation of image segmentation according to BP neural network, fuzzy BP neural network and the improved method. Evaluating the result through the following index:UM, GC, UMA, time. Make the contrast of result from three methods proves that the improved method improve the accuracy, and cost less time. It is a feasible method for image segmentation.