The Research Of Pixel-level Multi-sensor Image Fusion Algorithms Based On DWT And ICA

Multi-sensor image fusion technology is a hot topic in the research fields such as image understanding, computer vision and so on. It's widely used in the fields of target identification, intelligent robotics, and medical image processing etc. It has important impact for the national defense and security as well as economic development. Therefore the research of image fusion technology has important significance.In order to solve the issues of existed fusion algorithms which are easily affected by noise and no considering the image natural characteristics and so on, the pixel-level multi-sensor image fusion algorithms are deeply studied based on the discrete wavelet transform (DWT) and independent component analysis (ICA) theory in this thesis. Through a great deal of image fusion experiments, the thesis acquired a series of valuable results which can be summarized in the following aspects:1. Based on the analysis of wavelet theory, the fusion performance, advantages and disadvantages, and the possibility of improving of the existed image fusion algorithms based on DWT are deeply analyzed. Taking into account that it's difficult to obtain optimal quality standard reference images in practical engineering applications, inspired by reverse thinking, starting at constructed the worse quality of the reference image from the source images, a multi-focus image fusion algorithm which selects fusion coefficients according to the distance characteristics is proposed. The algorithm makes up for the shortcomings of traditional wavelet image fusion algorithms, which is fused in accordance with a statistical maximum value of the image data itself, by paying more attention to select useful information under the conditions of having reference images, thus enhances the effect of proposed fusion algorithm.2. By using the trait that independent component analysis (ICA) theory can guarantee the reducing of possibility of product redundant or false information in the process of fusion, which would help improve the fault-tolerant and robust of fusion algorithm, according to the link between matrix sparsity and ICA coefficients representation, an image fusion algorithm based on the matrix sparsity in ICA domain is presented, which has better anti-noise ability. The thesis has done useful try on the research of using blind source separation theory to image fusion algorithm.3. On the basis of the research of DWT and ICA, combination its merits, and taking fully into account the natural characteristics of images, an improved multi-sensor image fusion algorithm based on the ICA mixing matrix in DWT domain is proposed. It calculates the weights in the process of coefficients fusion according to the mixing matrix derived from ICA decomposition, and uses the weighted average method for the selection of fusion coefficient. The algorithm is more in line with the image natural characteristics, and fusion effect has been remarkably improved.