A Rbf Neural Network-based Image Processing Methods

In the past few years, image processing has been the focus and hot topic of researches. Additionally the digital image processing, it occupies a very important position in military, public security and so on. Digital images are often corrupted by noise during image acquisition and transmission due to a number of nonidealities encountered in image sensors and communication channels.There are so many filtering methods, such as The Standard Median Filter, The Edge-Detecting Median Filter, The Multi-States Median Filter, The Progressive Switching Median Filter, etc. But these traditional methods are inefficient in solving many problems. Therefore, as far as possible to find the filter noise, and protect the image detail while reducing the computation and speed up the computation speed of the filtering method of great significance, artificial neural networks remarkable high speed parallel computing ability and the auto-adapted training and learning capability definitely might satisfy the above request.Traditional noise filtering methods will lead to the degradation of the image details,1, This thesis gives an introduction to the basic elements of a neural network, the structural characteristics of network topology and the working properties under various network topological structures.2, Besides, it gives comprehensive accounts on the network topological structures and algorithm of the three widely-used neural networks of BP, RBF and Hopfield, points out its development actuality. Then one conceive of artificial neural network-based filtering method has been proposed in this dissertation. It is based on the RBF neural network. We can make a noise detector by a RBF neural network.The new restoration method is a filter combined with a RBF neural network noise detector and a median filter, in which detector the noise first so that the image details have been projected.3, Take a non-linear function as the example, carries on the Recurrent confirmation to this function, obtains works as when the probability sample number value scope is 300-400, the RBF neural network may the high accuracy approach this function and the computation load moderate conclusion.