Research On Using BP Network To Realize Video Clearing

In recent years, with the rapid development of monitoring technology, video surveillance has been more and more widely used. However, due to a variety of reasons, the received video image may be fuzzed by different factors, which will increase the difficulties when we use video images to provide accurate information. From the point of view, traditional clear methods have been mostly single-frame images, such as image enhancement, image restoration, etc. Though these methods have greatly improved the quality of the images, there are many drawbacks. Therefore, the search for a clear and effective method becomes particularly important.This paper considered the link between video frames. Due to the continuity of the movement of objects that if we link the position coordinates of an object in each frame, we can get the law of motion of objects in theory. When determining the location of the macro block coordinates, the motion vector is the most critical factor. Therefore, this article attempts to find a clear method to predict fuzzy frame's information through forward and afterward clear frames, by simulating motion variation of the inter-frame which is also called the relationship curve between position coordinates and motion vectors.As the BP network has greater advantages in analogizing both linear and nonlinear curves, and when apply it to the system modeling and identification, it can not only be exempted from the restrictions of system model, but also easy to achieve learning algorithm. Therefore, this article uses BP network to predict variation between the video frames, and to predict information of fuzzy frames through clear and relevant frame's information, so that to achieve fuzzy video clarity. In the actual research process, the size of block will have significant impact on the accuracy of the study and network training time. Therefore, it is necessary to divide the entire frame into some image blocks for processing. After a comprehensive consideration, this paper uses 4×4 as the basic object of study.In the course of the further study, we find that, because BP network is a local optimization method, it may appear into a local minimum point in the application process. At the same time, it may cause the network training time prolong too much and converge too slowly and some other shortcomings due to affection of the initial weight values, so in the later part of study we need to optimize the value of the network. By reading the relevant literature, we find that genetic algorithm is a global optimization algorithm, which can overcome the shortcoming of BP network, so if we use it in conjunction with the BP network, it is sure to play a complementary effect. Before deciding to use BP network to realize clearing, we firstly use genetic algorithm to optimize the initial weights of BP network, thus reducing the network training time and speed up the convergence speed. By comparing BP network with the traditional methods, it can be found that the BP network is more effective.At the same time, this paper designs a platform for users to use clearing method conveniently.This paper briefly describes the existing image enhancement and image restoration methods, and compares the adopted method with them. As the experimental simulation results shows, the methods used can overcome the traditional lack of clarity. At the same time, it can improve the video image of the quality effectively. Compared to traditional methods, this method can improve the peak signal to noise ratio of the image, reduce the minimum mean-square error, and produce a better visual effect. By using genetic algorithm to optimize the BP network's weights, it can shorten training time, speed up the convergence rate, and get a better clarity of results.