Study Of Mathematical Morphology Based On Neural Network And Its Application

On the base of a systematical analysis about the theory of mathematical morphology and neural network, we present that the two methods have many complementary aspects: Mathematical morphology has good filtering characteristic, but has not good adaptability and study ability from samples; neural network has outstanding properties of self-organizing and self-adapting, but has not good filtering characteristic of mathematical morphology. As a result, in the paper we present a method of mathematical morphology image processing based on neural network, which makes some improvement on original morphological image processing.The following is the main achievement in this paper:First, to binary image, we proof there are some relation between morphological erosion, dilation and perception model, and generalize the original binary erosive and dilative operation. As a result, we present the concept of degree morphological operation and establish the binary morphological model based perceptional structure.Second, we present the neural network model of grayscale morphological operation and educe learning arithmetic to gear the numerical value of structuring element.Third, we implement the original morphological arithmetic and mathematical morphological arithmetic based on neural network with computer program, and then we use the two methods to image noise filtering and edge extraction. According to the processing result, we draw the conclusion that the new method is better than the original mathematical morphological method.