Research On Image Restoration Method Based On Neural Network

Image restoration is an important research direction in the field of image processing.Its goal is to deal with degraded images so that it tends to be a clear image.When the image is acquired,the quality of the image is degraded by some influence.These influence includes image blurring of noise such as optical systems and motion,as well as noise from circuit.Image restoration is carried out by a variety of traditional methods such as inverse filtering,Wiener filtering,least squares method and so on.But the traditional method often has some problems in solving the problem of function approximation.Artificial neural network(ANN)is a popular mathematical tool,because of its own robustness and adaptive learning ability has been widely used.In this paper,the ANN is used for the image restoration,focusing on the use of BP neural network as a tool to restore the image.The BP neural network is essentially a mapping from input to output,which can learn a large number of input-to-output mappings without explicitly expressing the relationship between them.Only need to train BP network with the known model,the network will have the mapping capability between input and output.So the ANN can be used in the suitable that the Point Spread Function(SPF)is difficult or impossible to get.The main works are as fallows.Firstly,after the training sample selection,in order to reduce the input dimension of the network,the sliding window extraction feature is adopted.Considering the difference of image edge region and flat area degradation,the sobel operator is used to extract the edge of the image.The flat area of the image can be obtained by calculating the flat area and the edge area of the image.The edge part and the smooth part in an image need two network to train.After the output of the network is obtained,the result of the output is reconstructed to obtain the restored image.Secondly,explore the performance of different learning methods and different transfer functions in this paper's context.And through the experimental comparison to find a suitable learning method and transfer functions of this paper.Thirdly,in order to select the initial value of the BP neural network,a genetic algorithm is introduced.Through the introduction of the genetic algorithm and the principle of genetic algorithm to optimize the BP neural network,the genetic algorithm optimization BP neural network is proved to be the positive effect of the algorithm of this paper.Through the above main work,this paper has a better program for image restoration,in a more reasonable amount of computation and time-consuming circumstances,This program get a better effect.