Three-dimensional filters for multiview stereoscopic applications using layered depth images

This thesis proposes an extension of two-dimensional (2D) spatial filtering into three-dimensions for multiview stereoscopic applications using the layered depth image (LDI) representation. The proposed filtering scheme takes advantage of the depth information available when an image is represented with layers, and can return results that are comparable to or better than 2D filtering techniques for smoothing or sharpening stereoscopic images. In addition, the proposed filtering scheme is more efficient for multiview stereoscopic applications using LDIs than conventional 2D filtering since the filter needs to be applied only once for n views, whereas 2D filtering requires each view to be filtered separately (increasing computation time).

The proposed filtering method for smoothing stereoscopic images was also subjectively evaluated in a study involving 15 people. The results from this study indicated that the proposed filtering scheme received similar scores for both viewer comfort and naturalness when compared to the 2D bilateral filter.