Registration Of Visible And Infrared Images Based On Visual Attention

Compared with single-sensor, multi-sensor can provide more information. Multi-sensor complementary information is used together to get more rich, full and particular information. The fusion of the complementary information can help to improve our understanding of the scene and the performance of the system, so the more reliable results are obtained. The infrared and visible images are the most commonly used images. Fusion infrared image with visible image can help image understanding, targets detection and recognition and targets location etc. However, it is require strict registration on the geometry and gray scale before the infrared (IR) and visible (VIS) image are used to data extraction and analysis. So we study the registration method of IR and VIS images in this paper.On the basis of the existing methods of image registration, after analyzing the characteristics of the IR and VIS images, the difference the imaging mechanism, we focus on the registration method based on salient features, such as SIFT, SURF and BRISK. Their implementation processes are introduced. These three operators are used to register IR and VIS images, then the performance of three operators are compared according to the experiment of repetition rate and matching efficiency.Visual attention can help people lock significant target accurately and fast that avoids expensive computation in redundant information and improve computation efficient. When visual attention is applied to image registration, the interested area, the most concerned and the most effective information, can be searched quickly, then the basis for image registration is provided. The registration method of IR and VIS images based on visual attention is proposed in this paper. Firstly, the saliency map is obtained by using visual attention model proposed by Itti. Then WTA (winner-take-all) and IOR (inhibition-of-return) are used to obtain visual attention areas. After’Canny’operator is used to transform visual attention area to edge map, the invariant moment and correlation are used as criterion of rough match. After the rough match is obtained, the RANSAC algorithm is used to remove the wrong matched pairs. Then the parameters of transformation model are computed and image registration is achieved. Experimental results show that the registration algorithm is accurate, efficient and stable, and has good registration result.