Study On Image Fusion Technology Based On FCM Algorithm

With the development of space industry and military national defense, it becomemore and more convenient to get the remote sensing images. While the remote sensingimages exist such problem that the number of them is many enough, but the usefulinformation of them is little, and the resolution of them is low. So the image fusiontechnology in the field of remote sensing information speeds up. Image fusiontechnology is the method to deal with the image data which comes from the differentsensors about the same object synthetically, in order to obtain a new image that includesmore abundant information, more obvious features, and more conducive to targetrecognition for us. At present, the studies of image fusion have many enough, but mostof them are applied in the specific areas. And the selection of the concrete methodsalmost bases on the aim of the fusion, which lead to the case that the methods aboutimage fusion are in varied forms, without a unified mathematical model. According tothe status quo of image fusion, a method that combines the pattern recognition with datafusion is put forward to search for a general model for image fusion in this paper. In thisnew algorithm, firstly, classifying the image information generally and then using amethod of data fusion for image data to get the new data of the fusion image, achievingthe consistency of the fused image.The research object of the paper is the pixel level image fusion technology which isthe lowest level of the image fusion. It is a method operating in the space domain for thegray of every pixel in the image. Use the fuzzy C-average clustering algorithm (FCM)accomplish the clustering of every gray from0to255to obtain the probability aimingto the area of the target, the background and the uncertain separately. All the probabilitycan form a matrix which just the membership matrix about each gray value relative tothe three areas. Convert the membership matrix to the approval rating of threeincompatible propositions which is set in advance when they is regard to be sure of D-S evidence theory, and fuse the matrix from the different image about the them objectby D-S evidence theory in order to obtain the new data for the new fusion image.Considering that the FCM algorithm is sensitive to initial value and the robustness of itis very poor, the solution that the initial value should be changed for many times mustdo for a good clustering effect. In order to solve the defect thoroughly, a methodcombing the particle swarm optimization (PSO) that has a good global search abilitywith taboo search (TS) algorithm which has the ability to jump out of local optima isused in the paper to get the best initial values and proved to have better clusteringeffects at the aspects of the information entropy, average gradient, signal-to-noise ratio.In this paper, the following aspects will be involved in:The relative concept and the research status of pixel level image fusion will beintroduced, and the using algorithms in the paper such as fuzzy C-average clustering,particle swarm optimization, taboo search, and D-S evidence fusion algorithm will also be introduced in detail. The thought and the result of the theme that the fusion methodbased on FCM algorithm will be implemented by two images by the different sensorsabout the same object in the environment of VC, and then the final result about thefusion image will be analysis and evaluated.