Research And Application Of Image Registration Technology In Panoramic Mosaic

In many scenes of this information age,the images often carry more information than words.Comparing with words,images can make a deep impression on people because they have the iconicity and the vitality during the process of description same thing.Besides,people’s preferences for visual beauty make they can quickly understand the moral through image,this owns a higher efficiency than the words,so more and more people tend to using the image to express their meaning,Therefore,people’s requirements of image quality is increasing day by day,in the fields of panoramic mosaic,robot navigation,Remote sensing photography and so on,panoramic images of wide view and clear quality become the research and application needs.panoramic mosaic including two major steps,they are image registration and image fusion.To output the clear and no ghosting panorama efficiently,we should ensure the registration algorithm has high efficiency and good match result,thus registration technology is a key step in image mosaic.Image registration is the key technology of panorama stitching,the efficiency and accuracy of panorama stitching will be directly affected by the registration technology,so registration technology is very important in panorama stitching.Registration method based on the characteristics is widely used,Among this method,SIFT algorithm is a classical algorithm,its Registration results are more accurate and less affected by factors such as illumination,rotation and so on.However,it has the shortcomings of high complexity and low efficiency.So improvement research on this registration stitching algorithm has important theoretical significance and application significance.this paper based on the study of previous researches on the SIFT algorithm,analyzed the existing algorithms,found some spots to be improved,and put forward the corresponding solutions.The main work of this paper is as follows:(1)Image edge feature point extraction.Image edge carries a lot of feature points,however,SIFT algorithm needs to use Gauss smooth operation in the process of extracting the feature points,this will make the edge become very smooth,so the edge feature points extraction rate is not high,it will reduce the image matching accuracy,so we bringing in the Laplacian to preprocess the reference image and the image to be registered,in order to highlight its edge and increase the feature points of the images edge.(2)SIFT feature vector dimension extraction.During the process of feature matching,SIFT algorithm generates a high dimensional feature vector,it is time-consuming and effects the efficiency,So introduced unit information projection entropy to optimization,we use the unit entropy vector to describe the image,to speculate the difference between images by the distance between the unit entropy vector.(3)Matching point pair optimization.In the matching stage,due to the image size,light,noise and other effects,there is a part of false matching points,SIFT algorithm use random sample consensus algorithm to eliminate the error matching points,but it has a low efficiency.This paper used the improved random sample consensus algorithm to eliminate error matching points.The improved algorithm is applied to the panoramic image mosaic,and the image is fused with the weighted average method.Experiments show that compared with the original SIFT algorithm,the optimization algorithm can effectively improve the efficiency of the algorithm,reduce the error,and achieve a better matching effect.