Video Object Inpainting of Complex Scenes

A computer vision technique is proposed to remove a moving object in a video sequence for digital post-processing. We propose an automatic video inpainting algorithm which relies on the optimisation of a global, patch-based function. Our algorithm propose a novel motion inpainting technique for the completion of dynamic backgrounds. The empty region of the object in a frame is completed by filling in the dynamic background from the mixture components of background model. We successfully remove a foreground object and complete the unknown background. Foreground segmentation is done using spatial-temporal multi-resolution technique.The definition of edges is very important factor considered for video inpainting.Thus once the foreground object to be segmented is removed a hole is created in that particlar region. This hole is then filled by patch match process. For filling of holes Approximate Nearest Neighbour (ANN) search is performed. Were,the algorithm searches for nearly a approximate patch to fill in the hole region. This process is done simultaneously in all frames by comparing the foreground segmented images as references. The hole must be filled in a visually plausible way. Motion estimation and realignment is done using affine transform.