Research And Implementation Of Image Mosaic Method Based On Improved SURF Algorithm

With the continuous upgrading of shooting equipment,people can get high definition photos and rich image details.Based on the traditional image Mosaic technology,a large number of feature points will be detected when processing this kind of image with HD image quality,which affects the efficiency of image Mosaic.This thesis studied SIFT and SURF algorithms,compared the performance differences of the two algorithms,and selected the SURF feature algorithm with better performance for image Mosaic.Based on SURF algorithm,some shortages of this technology,such as stitching speed is a bit slower,the feature matching algorithm is not accurate enough and the image mosaic gap is produced by the distance-weight algorithm are improved.The main work of this thesis can be summarized in the following aspects:(1)Demonstrate the feasibility of preprocessing with local mean.In this thesis,the feasibility of image preprocessing by means filtering and its influence on reducing the number of feature points and program running time are demonstrated by experiments.It is proved that the mean filter of template size of 7×7 can be used for preprocessing.Considering that the image to be splicing has the characteristics of overlapping regions,the mean filtering can be limited to the local region,but the selected region is too large to make the effect not obvious,and too small will make the number of feature matches too small,so how to obtain the area threshold is a problem to be solved.The regional threshold experiment counted the distribution of feature points in 59 groups of images,and confirmed the rationality of the area threshold value of 0.6.By dividing the threshold region and limiting the mean filtering to the local region,the running time of the algorithm will be effectively reduced.(2)Propose a slope threshold feature matching algorithm to improve the accuracy of feature matching.In the phase of image feature matching,after image feature matching with the improved KNN algorithm,there are still some wrong feature matching,and the higher matching error rate will lead to a larger time cost for RANSAC algorithm to calculate.In order to solve this problem,an improved feature matching algorithm is proposed in this thesis.The slope of the corresponding feature matching pair in the two images is counted,and the slope threshold is calculated.The initial matching features of the improved KNN algorithm are further screened.Experiments show that the improved algorithm can reduce error matching and improve the accuracy of matching.(3)Eliminate the patchwork produced by the distance-weight algorithm.In the image fusion stage,the distance weight algorithm is used to fuse the image.In view of the brightness difference between the two images,there will be obvious splicing in the fused image.In this thesis,the improved distance weight algorithm is used to segment the image area and basically eliminate the splicing gap.After comprehensive experiments,the improved algorithm is improved in speed,accuracy and image detail.