Abstract: We propose a registration method for infrared and visible images based on shape context and histogram of oriented gradient (HOG) feature to overcome the limitations of single-mode image information. On the basis of foreground detection by Gaussian mixture model, we realize the contour feature matching using the proposed shape context and HOG feature. Matching is extended to the whole shape through a thin plate spline (TPS) transformation model. Then, we use the regularization and scaling characteristics to reorganize the corresponding relationship and estimate the transformation in order to reduce the estimation error. Finally, the random sample consensus (RANSAC) algorithm is used to remove the error matching points. Compared with existing shape context methods, this method combines edge and contour feature information with lower error and better robustness.