Research On Edge Detection Method Based On Image Feature Vectors And Neural Network

Digital image edge detection is one of the important paces in the image process. The edge includes a lot of useful imformation for the following work. The accurace of the edge detection influence the image process derectly.In this paper we mainly research digital image edge detection method. We use the image features and the BP neural network to establish an edge detect algrithm which has the anti-noise aility, and so on further studies for the image division completes the upholstery.First we introduced the related elementary knowledge about edge detection, including the earlier period preparatory work for digital image processing, such as image sampling and quantification, image smooth processing and so on. We have also carried on the contrastive analysis for the classical edge detection algorithms. And we illuminated each one's merit and insufficient to make that the the foundation for the new edge detection method. At the same time, the foundation part also introduced the edge detection algorithm appraisal method, which provided the scientific reason for the algorithm research.We then make the thoroughly research on the neural network edg detection algorithm. In allusion to the reality that there is too large inputing number for the network learning, we use the image feature vector to extraction information of the image to reduce the input scale. We improve the feature vector to make it more anti-noise ability. At the same time we use BP neural network to realize the process. The results of the simulation experiment show that the method is quite effective and robust than the classical methods.To speed up the rapidity and the accuracy of the new algorithm, we attempted many kinds of methods to improve the entire performance of the algorithm. In order to avoid the shortcoming of the BP neural network's easily falling into the local minimum, we use the particle swarm optimization algorithm to optimize globally, then use the BP network optimize locally. The experiment indicated that these methods greatly raised the speed of the algorithm and increased its reliability.