Smoothing Depth Maps for Improved Stereoscopic Image Quality

A technique to improve stereoscopic image quality of pictures generated from depth maps is proposed. Generally, there are two key problems with methods that render new images from an original image and its corresponding depth map: the depth map could contain artifacts (e.g., blockiness or contouring) and there is no information on newly exposed (disoccluded) regions. For the former, the solution is to acquire depth maps with better “resolution.” For the latter, a common solution is to fill the newly exposed regions with the weighted average of luminance and chrominance values of neighbouring pixels. However, this latter solution often leads to visible and annoying artifacts at edges of objects. We propose to smoothen depth maps before the rendering of new views so as to minimize the effects of blockiness and contouring as well as to generate gentle depth transitions at object boundaries to alleviate problems with disoccluded areas. Experimental results from subjective assessments demonstrate the effectiveness of smoothing depth maps in improving the perceived quality of stereoscopic images.