Adaptive video enhancement using neural network

Hyung-Seung Lee, Rae-Hong Park, Sunghee Kim

Abstract: This paper proposes an adaptive video enhancement method for digitally converted analog video. Analog video often has cross-luma artifacts and blurring artifacts by incorrect separation of a composite video signal. Even digital televisions suffer from these artifacts if high definition contents are converted from composite video contents. In order to reduce these artifacts, we trace signal patterns of artifacts and suppress them by using an adaptive linear filter. Moreover, to restore blurred edges of video, we adopt a neural network filter, in which weight coefficients are trained with artifact-free video. Experiments using a number of video sequences show the effectiveness of the proposed algorithm.