Intensity-Based Registration of Medical Images Using Penalized Maximum Likelihood

Abstract

We present a registration method in which we use the penalized maximum likelihood (PML) function, as a new measure, defined by the transition probabilities between the image intensities of corresponding pixels in both images. The value of measure is computed from the joint histogram obtained from the intensities of all pixel pairs in the overlapping area of two images and if two images are geometrically aligned, it is probably assumed to have a maximum value. By employing the PML function, we can assign much more weights on transition probabilities which occur in an important overlapping range. Therefore, the proposed registration method will provide a more accurate registration while being more robust to the various degradation environments. The accuracy and robustness of the proposed registration method as well as two other methods such as the mutual information (MI) technique or the maximum likelihood (ML) method are tested on real images. The experimental results show that our approach is more optimal registration method than other methods.