Research, Image Registration And Fusion Method Based On Partial Differential Equations

Image registration and image fusion are very important in pattern recognition and computer vision, which can deal with multi-source data, the analysis of sequence images, object detection and object recognition. They are wildly used in military, remote sensing, medicine, computer vision, etc. Image registration is a process which can calibrate the images that are from different perspectives, different time and different sensors and determine the geometry transform parameter. According to the fusion rules, image fusion technology can fuse complementary information. Through the image fusion algorithms, we can finally describe the consistency of different scene. Moreover, the final fusion results contain more abundant information, and reflect more accurate reality of the objects. For the local adaptability and flexibility, partial differential equation becomes an important research direction in each field of image processing. Therefore,this paper studies the theories and methods of image registration and image fusion based on partial differential equation.Our work mainly includes the following parts:(1) This paper studies the basic principle of the optical flow field and the registration model based on the optical flow field, According to the assumption of the constant of image gray and gradient, we combine the optical flow field and partial differential equation together and propose a modified model based on optical flow field for image registration.(2) This paper studies the basic principles of the mutual information. According to the optical flow field model, we introduce the mutual information as a new regulation term in a classical image registration model to get better results.(3) We introduce the definition of subjective contrast, and the model based on subjective contrast in the image fusion model.Then,we propose the model which is based on the gradient square.(4) We introduce the basic principle of TV model and the corresponding modified models. Then, we propose a new modified total variation based on the gradient square. Comparisons with other approaches demonstrate the superior performance of the proposed algorithm.In the experiment section, We compare the proposed algorithms with state-of-the-art on many differrnt types of images. The experimental results show that proposed registration algorithms and fusion algorithms all can get better results.