Quality-efficient universal demosaicing for arbitrary color filter array videos using spatial and temporal correlations

To save the hardware cost, most modern digital video cameras employ the single-sensor technology together with the red-green-blue (RGB) color filter array (CFA) structure to capture real-world scenes. Each pixel in the captured video is only composed of one primary color. This kind of videos is usually referred to be the CFA video (or mosaic video). For the purpose of visual display, it is necessary to convert the captured CFA video into the full-color RGB video. In this paper, we propose a novel quality-efficient universal demosaicing for arbitrary CFA videos. In the proposed method, both the spatial and the temporal correlations among the CFA data are taken into account during the demosaicing. Then an effective least square-based approach is developed to fuse the spatial-based demosaiced result and the temporal-based one. Experimental results on 27 test videos with ten typical RGB CFA structures show that the demosaiced videos generated by the proposed universal demosaicing method yield better quality in terms of average color peak signal-to-noise ratio (CPSNR) and demonstrate less color artifacts in terms of visual perception in comparison with two state-of-the-art universal demosaicing methods.