Studies On Image Registration Technique

Registration is a fundamental task in image processing used to match two or more pictures taken, for example, at different times, from different sensors, or from different viewpoints.The methods of image registration can be classified into two categories: the intensity-based matching approach and the feature-based matching approach. In this paper, we summarize the effort of image registration that the researchers have achieved, eventually puts forward my own algorithm.During the research that the intensity-based matching approaches, mostly study in the approach which is mutual information-based image registration. Mutual information-based image registration has been widely exploited due to its high-level automation and accuracy, especially in medical image registration. However, the object function suffers from the local extremum induced by interpolation, which obstructs optimization algorithms and leads to the wrong result of registration. In this paper, we propose a new interpolation algorithm to solve this problem. To improve the accuracy ,we provide some implemental issues,for example,the Powell searching algorithm,the choice of the gray level and consideration of outlier points. Experimental results indicate that the new algorithm can effectively avoid the local extremum and achieve better accuracy and robustness.During the research that the feature-based matching approaches, the concept of the monogenic phase is applied to feature point matching for infrared images and achieves the registration of infrared image. It evaluates the process of feature point matching from finding feature point by the Harris corner detector, forming a correlation matrix in term of the monogenic signal elements, then to the stage of RANSAC based on Sampson distance inliers selection so as to eliminate the false feature point pairs and to calculate the fundamental matrix of epipolar geometry.The simulation results show that the use of the monogenic phase has lead to nearly perfect matching.Lastly, the lack of the research in the paper is summarized, and then looks forward to the futurity of the research of image registration.