Image Inpainting Based On Multiple View Geometry

Digital image and video inpainting refers to the restoration of damaged portions in images or videos, so that the inpainted region does not stand out with respect to its surroundings. Currently, several algorithms using local information for single-view inpainting has been proposed. In this paper, we are interested in the inpainting problem where the missing regions are so large that local inpainting methods fail. As an alternative to the local principle, we make use of other images with related global information to enable a reasonable inpainting, which is based on the multiple view geometry.Homographic constraint is employed for planar object inpainting, the source region of candidate image can be transformed into current view by Homography, thus the target region can be inpainted using the source image. The experiments show that the image inpainting algorithm based on Homography performs well when the target region is large.For the damaged area of3-D (non-planar) object inpainting in the target image. We proposed a new inpainting algorithm based on epipolar geometry. First, the priority of each pixel on the "fill front" should be calculated, and the pixel with the highest priority is obtained. Then, the best exemplar pixel corresponding to the patch with the highest priority is searched along the epipolar line in the source image, and the relevant data is copied from the best exemplar pixel to the unfilled parts of the patch with the highest priority. Our experiments show that the speed and output quality of our algorithm compares favorably with recent patch-based algorithms.Finally, we proposed a marker hiding method based on Homography in augmented reality applications. The method presented here can automatically detect the target area then hide the marker, and also the restored region appears in harmony with the surrounding environment. The experiments show that our method demonstrates satisfactory visual effects and that shows high efficiency as well.