Sharpness Enhancement of Stereo Images Using a Depth-Based Per-Pixel Regularization

Abstract

Blur is one of the causes of visual discomfort in stereopsis. The application of 2D image sharpening algorithms to the left and right view can produce an interdifference which causes eyestrain and visual fatigue for the viewer. Additionally, it has been shown, through subjective tests, that the perception of sharpness is affected by depth. A 3D sharpness enhancement method for stereo images that incorporates binocular vision cues as well as depth information is presented. The proposed algorithm decomposes each of the input stereo images into a base and a detail layer. The visibility thresholds maps given by the Binocular Just Noticeable Difference (BJND), which include binocular mechanisms such as luminance and contrast masking are used as guidance maps in the computation of the base layer; additionally, the depth of objects in the scene is used to provide a per-pixel depth-weighted regularization to the computation of the base layer. The detail layer, defined using the input images and the computed base layer is boosted and then added to the base layer to provide the final sharpness enhanced images. The proposed sharpness enhancement method results in a low interdifference error of corresponding positions of the stereo pair and in an enhanced subjective visual quality. Comparative quantitative results in terms of interdifference using a publicly available dataset show that the proposed algorithm outperforms state-of-the-art algorithms. Qualitative and subjective evaluation results are also included in order to show the perceived visual quality improvement provided by the proposed algorithm.