New reconstruction method for intermediate views from multiple views

Reconstruction of intermediate view from arbitrary view plays an important role in the three-dimensional display system and the multi-view video compression. In this paper, we propose a new algorithm to create intermediate views from multiple views. First, a new centralized phase-based support-weights approach was proposed for disparity estimation, then the estimated disparity is refined by the multi-view information and the inpainting technique was employed to repair the remaining occluded regions. Our proposed algorithm could reconstruct the intermediate view with favourite quality, and the rectified images from the Middlebury database which contains the original intermediate images are used to test our algorithm. Experimental results show that our method can achieve a high image quality compared to stream-centric dense stereo method without introducing obvious artefacts.