Image registration method using adaptive nonlinear filter
K. Morishita, Tetsuo Yokoyama · Systems and Computers in Japan · 1988
Abstract The subtraction technique is applied widely in various fields. It performs the subtraction between images with different imaging conditions. In medicine, angiography called DSA is well known. However, it has a problem in the alignment of images to cope with the motion of the patient. The authors have developed a highly accurate alignment registration method to cope with such changes of position. This method is based on template matching and uses the adaptive filter in the preprocessing. However, the problem with template matching is that the accuracy of the positioning as well as the image quality is degraded in the region (flat region) where the gray‐level variation of the image is small. This paper proposes a system in which the boundary such as that of the vessel is emphasized in the pre‐processing, and the highly accurate registration is realized in the entire region of the image. To determine the filter which is effective in the preprocessing, a quantitative evaluation was made for the traditional differential filter and the frequency‐domain filter. The adaptive filter based on local statistics is applied, and it is shown that, compared with the traditional filter, the error caused by misregistration can be reduced to 1/2 ‐ 1/7. Then it is shown experimentally that the optimum window size (W) and the optimum emphasis factor (α) are W = 3 ‐ 7 and α ≥ 10 (5122 pixel size), respectively, and a high‐speed processing comparable to the speed of the differential filter can be realized.