Image Fusion And Application Of Dimensionality Reduction Technology

In this paper we first probe on the different level and mode of the image fusion based on the analysis of the current technical point of image fusion. We categorize the image fusion method into different levels and modes and choose some method to test, the experiment shows good result. We compare the fuse result by some subject evaluate rule (PSNR, RSME, D, RD, EN, CEN) and get some conclusion form it.Then we introduce the research actuality of the high-dimensional data theory, including the present of "the curse of dimension", the definition of dimension reduction, the method of dimension reduction and so on, we also probe on the linear and non-linear dimension reduction method (including PCA, PP, LLE, MDS and so on), through categorizing and summarizing the former result, we find the rule and the trend of the development.A local image fusion algorithm based on dimension reduction technology is proposed and used to fuse two gray images. This method combines image fusion technique with dimension reduction technology and divides images to be fused, then represents image blocks as related matrix, we use MDS (Multidimensional Scaling) method to process the related matrix to get the region of interest image, and then fuse the region of interest image with wavelet method so that the images to be fused are smaller than the former. The experiment results demonstrate that the proposed algorithm can improve the speed of processing and make a contribution to ATR and change detection process.As the communication bandwidth between sensors and a fusion center is limited, one needs to optimally precompress sensor out-puts-sensor measurements or sensor estimates before the sensors' transmission to obtain a constrained optimal estimation at the fusion center.A method which can estimate fusion coefficient by maximizing the local variance of the expectation fused image and estimate the expectation fused image by Optimal Weighted Least Squares (OWLS) based on wavelet transform, based on linear fusion model to optimally compress out-puts-sensor observations in given dimension is presented in this paper, This method can obtain a compressing matrix that minimizes the lost of estimation, so it is advantage to the transmission between sensors and the fusion center.