Research Of Multi-focus Image Fusion Based On Nonsubsampled Contourlet Transform And Pixel Relativity

The crucial part of image processing domain is image fusion technology.ln order to describe some scene or target accurately and completely,the images derived from the scene and target will integrate multiple by multi-sensor image fusion.Image fusion can make full use of the complementary information and redundant information in diverse source images,thus the fusion image is more reliable and less blurry,more over,the fusion image shows relatively optimal understandability and human visual perception,so the computer can detect,identify and classify the fusion image more easily.This dissertation summarizes and analyzes the basic concepts of the image fusion technology,and looks for new ways to enhance the fusion image quality effectively and retain more information of source images based on the research of mature image fusion algorithm.The chief innovations can be summarized as follows:1.On the basis of the level structure of image fusion,the methods and process of multi-source image fusion are summarized,the evaluation criteria and the principles of their application are discussed as well;Then the results of several typical transform domain image fusion technology are compared/and the transform domain image fusion technology based on multi-resolution analysis is discussed.2.Research the Non-subsampled Contourlet transform theory as well as multi-source image fusion,and then propose a new algorithm of image fusion based on Non-subsampled Contourlet transform.Decompose the images under multi-scale by Non-subsampled Contourlet transform,so the information of detail and edge can be obtained,and then select the corresponding fusion rules employed the obtained approach degree and regional characteristic of subband coefficients.The fusion results of this algorithm are better than traditional transform domain fusion algorithms on pixel level and space domain fusion algorithms.3.In the selection of fusion rules,the images are decomposed by Non-subsampled Contourlet transformed the fusion rules of the obtained high frequency subband coefficients and low frequency subband coefficients are different:selection at low-frequency subband coefficients, fusion rule based on improved pixel relativity algorithm is employed; selection at high-frequency subband coefficients, fusion rule based on extended spatial frequency and selecting large absolute of coefficients is employed.4. The experimental results show that the proposed fusion rule of this dissertation is effective and better than some traditional fusion rule in fusion ability.