Research On An Improved Algorithm For Infrared Image Registration And Stitching

In frared image processing technology is a very important subject in digital image processing, and has high application value in the field of land and resources investigation, exploration, environmental monitoring, battlefield surveillance, emergency safeguard etc. Especially in military field, a series of infrared reconnaissance images are used to form a complete image, which is high resolution and of large field of view without a gap, thereby realizing to weak infrared target search and tracking, infrared early warning, target recognition and detection, panoramic scanning military strategy and tactics.This paper first introduces the research meaning and purpose, research status of infrared image and expounds the advantages and disadvantages of the imaging principle, the infrared image generating mechanism, characteristics, and its military application. It focuses on analysis in the factors and the infrared image noise of infrared image of the original image, evaluation criteria, and further studies pretreatment of these factors on the infrared image.Secondly, it studies on pretreatment technologies image denoising, image enhancement and image segmentation technology, infrared image obtained through experiments, which is more suitable for the process of the method and process of infrared image pre processing. On the image denoising, front either on spatial or level frequemn level. A hybrid filter can improve the treatment effect of infrared image Gauss noise and salt and pepper noise; On the image enhancement front the processing method. Help enhance the image but it at the same time increases the noise; On the image segmentation front the Otsu is more suitable for segmentation prosess of infrared image.Again, it sheds light on study in the point feature extraction algorithm. It focuses on the SIFT, SURF-two kinds of local invariant feature extraction operator. In this case, the contrast analysis is carried out by feature point experiments on their detection speed, descriptor generation speed infrared registration, comprehensive performance, introducing homography constraint RANSAC fine matching based algorithm to improve the image registration accuracy and reduce the false match rate image.Finally, it introduces the basic process of infrared image mosaic. After the registration of infrared image is completed we found out, the overlapped area which still exists gap or a clear transition trace phenomenon, and did further research in image mosaic algorithm. We finally analyze and summarize the image mosaic method which is commonly used, and according to the ghost image mosaic present the improved weighted average method, further validating its dependability and practicability by simulation experiments.