A Study On Algorithm Of Edge Detection Of Digital Image Based On LVQ Neural Networks

While the gray scale of two adjacent fields is obviously different we will think that there is an edge at the border of the two fields. Edge expresses the end of one field and the start of the other field. Edge detection is a course that processes and identifies the edge of image with different of algorithms. It is one important technology of digital image processing, and one important basic subject of signal and information processing. Edge detection has been extensively applied in a lot of fields, such as image processing, artificial intelligence, computer recognition, pattern recognition and classification, fault detection etc.The major contribution of this paper is summarized as follow: 1. The paper summarizes the principle and development ofedge detection, and systematically study on the basic theory of edge detection. And then proposes that edge detection can be seen as a kind of classific problem to study on.2. The paper analyses the research orientation of edge detection based on neural network, and explains the theory of edge detection based on neural network detailedly.3. Aimed at the defects of the traditional edge detection of anti-noise performance, the paper for the first time proposes the new algorithm based on LVQl (Learning Vector Quantization) neural network. The paper constructs the feature vector made of three feature scalars to extract valid information of image which is the import sequence of LVQl network. Then the simulation results are provided to show the algorithm not only can be used to detect the edge of image but has better anti-noise performance.4. Aimed at limitation of LVQl network of "dead" neuron and worse convergence performance, the paper proposes an advanced algorithm based on LVQ2 neural network. The structure of the LVQ2 network is similar to LVQl network. But LVQ2 introduces areasonable algorithm in study rule. The new algorithm learns both the first winning neuron and the second winning neuron. The speed of learning introduces the reasonable mode of variable step size. Then the simulation results are provided to show the algorithm makes continuity of the export edge image and anti-noise performance better.5. The paper systematically generalizes the advantages and disadvantages of both the two algorithms. In the end, the paper discusses and prospects the future study orientation.