Study On Infrared And Visible Image Registration

Infrared image is widely used for its strong capability of detecting hot targets. The hotobjects can be located after infrared and visible image fusion which is based on registration.So it is very important to study the accurate and efficient registration technology.Single modal image, such as infrared or visible image, is studied in this paper. First, theSpeeded-Up Robust Features (SURF) algorithm, which has strong robustness, is used todetect interest points in the test images. Second, the matching points are found by using therules of the minimum Euclidean distance. Then the ill-conditioned correspondences areeliminated by using the improved Random Sample Consensus (RANSAC) algorithm. Finally,the affine matrix is computed by the Least Squares Method, and the process is accomplishedby using the aforementioned matrix and bilinear interpolation. The accuracy of results canreach the sub-pixel level.For multimodal image, the infrared and visible image registration is mainly discussed andthe experiments are done from different aspects. For example, methods based on colorconversion model and inverse image are studied to solve the problem of the great differenceof intensity. For infrared image lacking of texture compared to visible image, research basedon Contourlet decomposition is made. Because the edges of the two images are consistent ona large scale, an approach based on edge and SURF algorithm is proposed, which reachesprecise accuracy. Images are obtained from different angles, and experimental results showthat the smaller the angle difference is, the higher the alignment accuracy can reach.