Improvement on optical flow based video deinterlacing by adopting Flow Vector And Intensity Reliabilities

A recently introduced optical flow based deinterlacing algorithm showed a promising performance for the sequences with well estimated optical flows. However, estimated optical flows mostly include some errors, which cause significant artifacts. To reduce such artifacts, we propose an improved algorithm by considering flow vector reliability in terms of motion linearity, motion uniqueness, and consistent motion in a flat region. Moreover, to maximize the resolution of deinterlaced images, we propose intensity reliability which utilizes the field parity information. Experimental results show that the proposed algorithm noticeably improves the quality of deinterlaced images.