Research On Cumulative Error Processing Method Based On Feedback Mechanism In Panoramic Mosaic Image

With the development of computer vision technology, panoramic images shows great application value in the real world. Image mosaic refers to register same scene and overlapping image sequences, and fuse a seamless panoramic image. The technology takes into account wide field of view and high resolution of the image. It has become a hot research topic in computer vision and virtual reality technology. And it is widely used in areas such as environmental monitoring, video surveillance, remote sensing image analysis, virtual reality, video data compression and transmission and other fields. In order to improve the quality of the mosaic image, this paper mainly studies the problem of accumulated error in mosaic panorama:(1)The paper introduces the background and significance and research situation of image mosaic, summarizes the basic theory knowledge and analyzes the key technologies and process of image mosaic. In addition, the paper compares their advantages and disadvantages of some mainstream algorithm of key technologies of image mosaic.(2)In image registration, aiming at the error matching of traditional SIFT algorithm, this paper proposes a SIFT image registration algorithm based on feedback mechanism. After extracting the feature points by SIFT, firstly, the algorithm according to the Euclidean distance between the feature points to eliminate the obvious one-to-many (many-to-one) false matching points. Secondly, rough matching is performed according to the optimal overlap ratio and the principle of the constraint mutual matching based on feedback mechanism, additional, rough matching can eliminate some false matching points. Finally, exact matching is performed by the RANSAC algorithm under the feedback mechanism, and the residual error matching points are eliminated. Experimental results show that the method has good adaptability and realizes from rough to exact image registration, and ensures the quality of feature points and the accuracy of transformation model parameters. It can effectively reduce the error caused by image registration.(3)In the panorama mosaic, In order to solve the problem of accumulated error in panoramic image mosaic, this paper proposes a strategy that transforms the reference image based on the multiple cylindrical projections. On the basis of analyzing the mosaic error, firstly, the algorithm improves the cylindrical projection algorithm by combining phase correlation and multi-resolution decomposition to reduce the amount of computation in the cylindrical projection transform. Secondly, the article adopts the strategy of hierarchical matching based on reference image transform to mosaic image sequence and projects the mosaic image of each layer on the cylinder, which avoids the error accumulation unidirectional in mosaic. Finally, the algorithm generates high-quality results and renders seamless output panoramas through the brightness adjustment algorithm based on variance and the Laplacian graded blending. The Experiments show that the strategy can effectively distribute the accumulated error of the adjacent image and improve the precision of panorama.(4)On the basis of the above theory, the paper realizes the seamless cylindrical panoramic image mosaic, and the qualitative and quantitative analysis of several kinds of image mosaic methods. Additional, the algorithm is evaluated from two aspects:the quality of the mosaic image and the execution time of the algorithm.