Research Of The Image Stitching Based On The Video Sequence

Image is one of the most important channels for human to obtain information.As time changes,people are no longer satisfied with common perspective images.Instead,they are seeking for wide perspective images.However,wide angle lenses,which can obtain wide perspective images,are relatively expensive,and there is a certain degree of distortion.Therefore composing wide angle image with general images from common devices becomes an active research.In recent years,plenty of image stitching algorithms based on the video sequence sprang up,but the stitching result is not sound ideal enough when the size,light and angle of the video sequences change.As a result,some improvement projects are proposed according to the problems of the existing image stitching algorithms.Some improvement strategies are detailed as follows:(1)Considering redundancy between video frames,the influence of the interval between selected frames on the number of matching points,the matching time and the error matching rate is statistically analyzed,and the key frame extraction method based on statistical curve is put forward,greatly reducing the workload of stitching.(2)Aiming at the problem of the low speed and low matching precision of SIFT,an improved SIFT matching algorithm combing multidimensional scaling and texture character is proposed.First,the computing speed of SIFT feature matching is improved through dimension reduction of 128 d SIFT feature descriptor with the MDS.Compared with others algorithms,MDS can ensure topological structure in geometry.Then an improved bidirectional matching strategy is proposed based on the consistency of characteristics and the consistency of ratio.Finally,the uniform rotation invariant LBP characters of the matching points' adjacent area on original gray image are analyzed to make further efforts to reduce the error matching rate.(3)The Poisson fusion is introduced into stitching procedure to remove mosaic seam in the images resulted from fusion.At first,the overlap region of the image sequences is obtained with SIFT image registration algorithm.Then,the parameters of Poisson fusion are initialized.Finally,the seamless splicing is realized by solving the Poisson fusion.In order to avoid of color bleeding in Poisson fusion,the Weighted Poisson Fusion(WPF)is proposed based on the traditional Poisson Fusion.The simulation experiments are carried out using the real images from different scenarios as instances.The experimental results show that the proposed method performs over other algorithms,which can provide better matching accuracy and fusion effect,as well as require shorter matching time.