Study Of Image Registration Method Based On Mutual Information

Image registration is the space transformation process of two or more images of the same scene taken at different times, from different viewpoints, and/or by different sensors. It plays a very important role in computer vision, pattern recognition, medical image analysis and remote sensing etc.Mutual information based image registration method does not require pre-process of the image, easy to implement different modes of automatic registration between images, but the computation is large. So this paper finds its parameter optimization of registration measure function by using combined global search operator PSO and local search operator Powell to improve the accuracy and efficiency.In addition, this paper proposes a new image registration method combined spatial gradient and mutual information, which takes into account the image intensity and spatial information. Firstly calculate the gradient amplitude and angle by using wavelet multi-scale product, then combine the spatial gradient and normalized mutual information to define a kind of new measure function. Through the simulation, the new method compared to the traditional normalized mutual information method has a lower sensitivity to change region and noise in images, the smoother curve with a sharp peak value, so the new method is easy for selecting the best registration parameter; moreover, the mutual information between image after registration and reference image by the new method is larger. These prove that the new method is more effective and more robust.