Virtual View Reconstruction from Stereoscopic Images

An efficient disparity estimation algorithm based on block error and smoothness error was proposed for reconstructing virtual views. The occluded regions were detected by measuring the reliability of disparity estimation, and the disparity values in these regions were determined using object-based interpolation. Finally, the left image and right image were projected onto the plane of virtual view to be reconstructed. This resulted in two projected images. The virtual view was created using a weighted average of these two projected images. The proposed method was tested with several image pairs. Experimental results show our method can provide good disparity map and the reconstructed virtual images have high image quality and good stable depth when viewed stereoscopically.