Robust Image Registration Using Selective Correlation Coefficient

\Ve propose a new method for robust image registrat,ion called 'Selective Correlation Coefficient (SCC)' in order to search images under illconditioned illuminat,ion or partial occlusion. A correlation mask image is generated for selecting pixels of a image before matching. The mask image can be derived from a binary-coded increment sign image defined from any object image and the template image. The masking rate of occluded regions is theoretically expected to be 50%, while unoccluded regions have much lower rate than 50%. We realize robustness under ill-conditioned environment since inconsistent brightness of occluded regions can be omitted by the masking operation. We propose, furthermore, the mask expanding procedure to get more stable robustness. The effectiveness of probabilistic masking increases by the procedure, resulting the rate around 70% of masking of occluded regions. This paper includes a theoretical modeling and analysis of the proposed method and some experimental results with real images.