Summary: | 博士 === 國立臺灣科技大學 === 資訊工程系 === 102 === Recently, compressing red-green-blue (RGB) color-filter-array(CFA) videos has been studied extensively, in which the structure conversion based and the demosaicking based compression schemes are mainstream. The former can deliver higher compression quality but has difficulties in converting irregular RGB-CFA structures, while the latter is applicable to arbitrary RGB-CFA videos but suffers from the RGB color deviation problem due to the chroma subsampling.
In this thesis, we present two novel RGB-CFA video compression schemes, where one is an improved demosaicking based compression scheme in H.264/AVC and the other is a novel CFA video compression scheme in high efficiency video coding (HEVC). First, to solve the RGB color deviation problem which occurs in the existing demosaicking based compression schemes in H.264/AVC, we propose a color deviation-free universal subsampling strategy for compressing arbitrary RGB-CFA videos in H.264/AVC.
Under the framework of Yang et al.'s universal chroma subsampling strategy, we replace retaining the original Y luma component with adopting a optimal modification for the Y luma component such that the RGB color deviation problem can be resolved. Next, to develop a CFA video compression scheme which is applicable in HEVC and avoid the problems of the existing structure conversion based and demosaicking based CFA video compression schemes in H.264/AVC, we propose a high quality universal intra coding scheme for compressing arbitrary RGB-CFA videos based on HEVC. The proposed scheme first demosaicks the neighboring reference pixels and then predicts, according to the CFA structure, the color value of the target pixel using the color values of the identical components in the reference pixels. The proposed scheme avoids the difficulty in converting irregular CFA structure and the quality degradation caused by chroma subsampling. Experimental results demonstrate that the proposed RGB-CFA video compression schemes achieve better quality of the reconstructed RGB-CFA videos under the same bitrate, when compared with the existing ones.
|