On Pixel-level Image Registration And Fusion Based On Wavelet

In recent years, image fusion has been an important and valuable technology in the Image Understanding and Computer Vision fields, which has been widely applied to geology, agriculture, mapping, computer vision, military remote sensing, medicine, biology and so on. The main research point of the image fusion technology is to integrate the image information which obtained by different kinds of sensors, and to improve the reliability and the observability of the image by processing the redundant datum and the complementary information among the images. Therefore, the fused image satisfies the perception of human and machine and will be propitious to image processing, such as object recognition, image analyzing, feature extraction and object trace.In this thesis, the pixel-level image registration and image fusion has been researched. The research involves image registration algorithm and image fusion algorithm based on the multi-Bourse image fusion pre-process. This thesis mainly consists of three parts as follows.Based on the comprehensive discussion and summarization of previous papers and research work, basic concepts, models, methods, techniques and applications of image data fusion, especially for pixel-level image fusion, are analyzed. Furthermore, image evaluation for the performance of image fusion is analyzed. Several evaluation criteria are presented and classified in qualitative and quantitative.It is well known that image registration is the basic of image fusion. An improved image registration method is proposed by comparing the resemblance of the original image and its approximation of wavelet decomposition in this thesis. In this method, four sub-images are used to be registered respectively for the first time. Then, Harris corner detector creates a feature point and the improved Hausdorff distance is used as the similarity measure for realizing image registration for the second time. Finally, we use the four sub-images that have been registered to reconstruct the result image. The experiment results show that this method can raise the accuracy and the precision of image registration.Based on the research of multi-resolution wavelet-based image fusion, an improved fusion algorithm on the wavelet transform is proposed. According to the property that the energy of image on the lower frequency sub-band and the details on the higher frequency sub-band, the high frequency band are fused by Weighting fusion based on and edge, then the lower frequency band are fused by comparisons between energy algorithm and Weighting algorithm. According to the analysis of the simulation image and datum, it is verified that the emergence of vagueness phenomenon can be avoided because of image average and much clearer image and much easier identification are yielded. The simulation results show that the image fusion based on wavelet transform has faster processing accuracy, better fusion effect, and clearer images.