Research Of Image Fusion Based On Average Gradient And Wavelet Multi-Resolution Analysis

Many intelligent systems for measure and control have been improved by the technology of multi-sensor information fusion, which has emerged as a new and promising research area. Being an efficient method of information fusion, image fusion has been used in many fields such as machine vision, medical diagnosis, military applications and remote sensing. How to make efficient image fusion for multi-source remote sensing image of the same object, make the best of the useful information, improve the system's ability of correct recognition, judgment and decision, is an important aspect of data fusion research. The goal of image fusion is to enhance the reliability of original image information and interpretation by incorporating different or complementary information and create new images that are more suitable for the purposes of human visual perception, object detection and target recognition. In the fusion image, most of all the image characteristics and information have been preserved, and the definition and the spatial quality of the original image have been improved.In order to adequately make use of all kinds of remote sensing images information, based on different feature of multi-sensor image and the advantages and disadvantages of classical image fusion methods, a new image fusion method based on average gradient and wavelet transform is the subject mater of this thesis.In this thesis,we introduce the theories of multi-resolution analysis and wavelet transform first, and then present some pretreatment technique,especially the wavelet denoising method based on soft threshold and the image registration method based on spatial projection transform. Finally, we propose the image fusion strategy based on average gradient and wavelet transform. In the wavelet fusion processing, the fused approximate coefficients are obtained with weighted average method. For the bigger average gradient of the each decomposed approximate coefficient, we choose a big power gene. The other approximate coefficient chooses a small one. The fused detailed coefficients are obtained by setting each coefficient equal to the corresponding input image wavelet coefficient that has the greatest average gradient. The experimental results show that both the fusion methods are more effective than the other methods proposed in this thesis.In fact, this thesis leaves much to be desired, and it is necessary to do further researches on image fusion in the future.