A Class Of Modified Hopfield Image Restoration Algorithm

The inherent parallelism and powerful computation capability of the Hopfield neural network make the image restoration algorithms based on this network com-monly possess fast running speed. However, on the one hand, the state of the neuron in the network can take only step values which lead to that the network can't precisely converge to the minimum of the network energy function, it affects the quality of the restored image. on the other hand, due to the lacking of se-lecting of the updating neurons, the quality of the final restored image can't be improved any longer. Hence, a lot of modified algorithms and models are proposed to overcome these deficiencies.The main work of this thesis is divided into three parts. First, by introducing the local fastest descending direction, this thesis proposed a kind of general update rule in the continuously changing Hopfield neural network(CHNN) model, this kind of general update rule makes the network under any updating modes guarantee that the energy is decreasing at each step. Then by researching the eliminate highest error(EHE) criterion, this thesis proposed a class of balanced EHE criterion in the CHNN model. The balanced EHE criterion not only improves the the probability that the state of the neuron translates to its correct state during the iterating process of the algorithm, but also decreases the searching space of the algorithm which improves the running speed of the algorithm in a certain degree. Finally, by combining the work of the two above aspects, this thesis proposed a class of modified Hopfield restoration algorithm. This thesis is composed of the following five chapters.In the first chapter, the background of image restoration and the main achieve-ments on the traditional image restoration and neural network image restoration are presented.The second chapter is devoted to introducing some correlative knowledge of image restoration model and Hopfield neural network model.In the third chapter, we introduce two classic Hopfield neural network models for image restoration.In the fourth chapter, we propose a kind of general update rule for guiding the design of kinds of image restoration algorithms in the CHNN model. At the meanwhile we proposes a class of balanced EHE criterion for guiding the selecting of updating neurons during the iterating process of the algorithm. At the end of this chapter, a class of modified Hopfield image restoration algorithm is proposed and the convergence of this algorithm is proved at the meanwhile.In the fifth chapter, the numerical experiments of the algorithm given in this thesis are implemented. Experimental results show that the proposed algorithm can get better restored image.