A Modular Neural Network For Image Restoration Research And The Optimization

This paper introduces the basic theory of image restoration,and some commonly used methods of image restoration are discussed,based on artificial neural network theory,the image restoration problem is abstracted to a regression problem and solved by neural network technology.The artificial neuron model is discussed as the basic unit of neural network,the most commonly used feed forward perception network.With the neural network as the research object,the BP learning algorithm and neural network learning methods are researched.Analysis of the defect of the neural network,and introducing the concept of modular neural network is discussed,where the modular neural network advantages,methods of analysis and design method of modular neural networks are given.The modular neural network subnet using improved genetic algorithm and particle swarm optimization algorithm based on step for image restoration,presents two improved algorithms based on BP neural network,and has carried on the image sub network before and after the improvement of recovery experiment based on.Aiming at the problem of single neural network in large size image restoration in the restoration effect is poor,the design of modular neural networks is used for image restoration.Firstly,according to the size of the sub module design sample image training sets and the input to output mapping relationship,then use modularization copy and decomposition,design reuse has sub module design and good modular neural network.The modular neural networks and single neural network the design good were compared using standard test image noise of the same size different training,from the network performance and analysis of two aspects of the effect of image restoration,the effectiveness of the proposed method.Finally the application of optimization algorithm of modular neural network subnet,enable the network to further improve the performance,image restoration effect is improved..