Object removal and inpainting in multi-view video sequences

An inpainting algorithm for multi-view video sequences, which fills in a mask area without visual artifacts, is proposed in this work. After a user specifies a mask area in a frame, the proposed algorithm clones it to the other frames in the input sequence using temporal and inter-view correlations. The proposed algorithm then selects sites along the mask boundary, over which source blocks are pasted. We develop a cost function, which measures the spatio-temporal consistency of an inpainted result, and minimize the cost function to select the best matching source blocks. Simulation results demonstrate that the proposed algorithm removes and inpaints objects within mask areas efficiently without causing blocking or flickering artifacts.