Artifact Reduction based Multi-view Generation Algorithm for Sparse Camera Configuration

A high quality multi-view generation algorithm is proposed to eliminate the camera parameter limits for a sparse camera configuration. The proposed algorithm includes multi-view generation and artifact reduction. The feature points of the foreground objects in the input images are used to find the disparity function for the central view and then we use this function to determine the disparity calibration function for all virtual views to replace the camera parameters. To improve the generated virtual image quality, artifact reduction based on a modified inpainting algorithm is proposed to reconstruct the holes and false contours. Experimental results show that the quality of the virtual view images in multi virtual view generation and in artifact reduction can be raised by 2.67dB and 4.57dB, respectively in terms of PSNR. From the experimental results, it is found that the proposed algorithm can provide multi virtual views of high quality and can be adapted to different multi-view display systems.