Research On Seamless Inlay Technology Of Panoramic Image

Image mosaic technology is an effect way to obtain a panoramic image, which solves the limitation of getting wide view angle and high resolution image for common camera so that human can express information more effectively, understand the real world better. As a hot research topic in the field of image processing, image mosaic technology is widely used in computer vision, medical image processing, virtual reality, aviation, military, geological exploration and other fields. In order to obtain high quality panoramic image, this paper studied panoramic image seamless mosaic technology, mainly including:Firstly, this paper described the research background and status of panoramic image mosaic technology, and deeply studied the basic theories of image mosaic. It also studied some key technologies, and compared and analyzed the advantages and disadvantages of some mainstream algorithms.Secondly, In view of the questions of producing seam-line and blurring for fusion algorithms based on spatial and frequency domain, this paper proposed a seamless image mosaic method based on gradient domain. It mainly researched and improved the algorithms in image registration and fusion. For image registration, it proposed Harris corner algorithm with adaptive threshold to solve a problem that threshold is not universal. Before image fusion, it used global brightness adjustment algorithm to eliminate exposure difference and realize smooth transition. For image fusion, it improved Poisson fusion. The key elements in fusion region were improved and adaptive mixing gradients were added to guide interpolation for fusion. Experimental results show that this method can effectively reduce the influence of exposure difference, eliminate the seam-line. While keeping the definition, it improves the quality of mosaic images and achieves the goal of seamless mosaic.Finally, Aiming the problem of multi-image mosaic, this paper proposed a method of cylindrical panoramic image mosaic based on SIFT feature. Considering the limitations in expressing the panorama for plane projection model, it used cylindrical projection model to transformed images into cylindrical coordinates to mosaic. On the problem of accumulative error, it put forward the middle frame as the final reference image, and a mosaic way of starting from both ends and taking turns to change the reference image. For feature extraction, it used SIFT algorithm for SIFT feature has invariance of image scale, rotation and illumination. After image matching, nonlinear optimization algorithm (L-M) was used to improve registration accuracy and further reduce the accumulative error. Experimental results show that this method can effectively reduce the accumulative error, realize panoramic seamless image mosaic and obtain a more precise panorama.