Medical Images Registration Method Integrating Features and Multiscale Information

Abstract

Medical image registration is the key technology to the modern medical images and it is the basic issue of medical image fusion. A medical image registration method integrating features and multiscale information is proposed in the paper. Firstly, the edge feature images obtained by Canny operator are registered by Principal Axes method. Secondly, Harris corner detection method was used to extract the corners of the couple images. Then the corners of the couple images are registered by Iterative Closest Point Algorithm (ICP). Finally, the couple medical images were decomposed by wavelet transform, the mutual information is selected as similar measure, and only the low-frequency sub-images were selected as registration objects. Then, the registration process is begun from the coarse scale until the finest scale by Powell algorithm. Thus the medical images registration can be achieved through the above registration procedure. Experiment results show that this method not only can achieve a higher registration accuracy, but also can reduce the computational time greatly. The accurate registration results also can be achieved in the noisy environment.