Image Fusion Based On Pulse Coupled Neural Network

Images are important in people’s life. One of main purposes of image processing is to get more effective information from images by analysis and disposal. Image fusion is an important part of image processing, and it aims to register image data of the same object from different sources and combine preponderant or complementary information of each image data to generate new image data. With the spread of image fusion technology, more and more fusion methods are proposed. Recently, methods based on pulse coupled neural network show their advantages gradually, and are concerned by a lot of people.This paper studies image fusion based on pulse coupled neural network. Firstly, this paper summarizes image fusion’s basic concepts, development status, evaluation index and some fusion methods. Secondly, this paper introduces mathematical theory knowledge of pulse coupled neural network and development of image fusion algorithm based on pulse coupled neural network. Then this paper presents two classical image fusion method based on pulse coupled neural network:image fusion algorithm based on dual-channel pulse coupled neural networks and algorithm based on spatial frequency-motivated pulse coupled neural networks in image fusion nonsubsampled contourlet transform domain.Against some shortage of image fusion method based on pulse coupled neural network at present, this paper proposes some improved ways for image fusion algorithm based on dual-channel pulse coupled neural networks and algorithm based on spatial frequency-motivated pulse coupled neural networks in nonsubsampled contourlet transform domain. At first, for dual-channel PCNN fusion method, this paper improves and simplifies pulse coupled neural network model, set linking coefficient by the characteristics of images and use the results of the last iteration to replace the balance factor of internal activity. Secondly, for nonsubsampled contourlet transform PCNN fusion method, this paper simplifies pulse coupled neural network model on a certain extent, and selects region-max rule when combined with nonsubsampled contourlet transform and linking coefficient is adjusted depending on the excitation. At last, this paper compares two kinds of improved fusion method with the original method and other fusion methods through experiments to analysis the advantages and disadvantages of each method.