Research On Image Registration Algorithm Based On Video Frame Sequence

Image registration techniques,as the basis for image processing,its practicality and accuracy of the algorithm will directly affects the success rate and speed of image processing.Therefore,in recent years,the study of image registration algorithm is the focus of the study at the image processing system.Image registration comes from many practical problems of actual areas,such as the difference of detection,the different conditions at different time,to obtain the three-dimensional image information in case of a scene change information acquisition;different sensor information obtained fused;image the pattern or object recognition.In this paper,the research question is image registration method of video frame sequences with weak redundancy.For global image registration methods,this paper summarizes the pixel-level image registration method,for more accurate registration requirements,this paper got the whole pixel offset frequency domain using phase correlation method,and then iterative Taylor series method,multimodal fitting extension phase correlation method to calculate the offset sub-pixel least squares template matching method based on the final registration results obtained sub-pixel accuracy,to meet the requirements of high-precision image registration.However,the situation for the angle of rotation of the actual existence of the surveillance video,zoom,pan,and other non-rigid transformation,this paper presents an approach based on affine transformation model is accurate to sub-pixel image registration method: after extracting SIFT feature points for image registration,the false matching points are eliminated initially based on RANSAC,and then which are removed further more by using inter image homography cross-correlation properties correlation and feature point.And then the affine model is calculated by the feature points estimated according to the vector independent guarantee choice principle between two feature points.Finally,the affine transformation model is corrected according to the corresponding relationship of the affine transformation parameters.The comparing test result shows that,the proposed algorithm can effectively avoid the problem of inaccurate registration results caused by the error affine transformation parameter,not only reduce calculation time significantly,but also achieve highly registration accuracy.Regional presence for the case of local motion video sequences,the local irregular sports area seriously affected the registration accuracy of the global registration method,background modeling method is susceptible to interference complementary global non-redundant information,it is difficult to accurately locate issues of local sports area,this paper presents a pinpoint local motion area approach: by combining video frame sequence information,first determine whether there is local motion between frames registration area,to avoid the existence of local motion regions affect the overall sub-pixel the registration accuracy,followed by the method given the global and local motion image registration coexistence of regional and local motion and global motion accurately separate sub-pixel registration.So that in the interference regions to obtain the local motion more accurate registration results.Provide useful information for the needs of the follow-up of super-resolution reconstruction.After get an accurate image registration results,this paper combines sinc function interpolation method,effectively use the complementary sequence of video frames between the information given and effective improve the image resolution.