Locally estimated heterogeneity property and its fuzzy filter application for deinterlacing

This paper presents an intra-field scanning format conversion method using two filters: bilinear filter (BF) and fuzzy-based weighted average filter (FWAF). The proposed method is intended for black and white images, luminance component of YIQ color space, or each color component of RGB color space. We start from the notion that pixels to be interpolated can be classified into two areas based on local variance: homogeneous and heterogeneous areas. According to the local variance criteria, we apply the FWAF to the heterogeneous area and the BF to the homogeneous one, producing good visual results. Our FWAF consists of an intensity similarity filter and a geometric closeness filter. The latter is used to populate the heterogeneous area with the missing lines, due to its high deinterlacing precision. Our experimental results show that the proposed approach provides satisfactory performances in terms of both objective metrics and visual image quality. We used parameter tuning on our training set to explore the relationship between objective quality and computational complexity. We report on how to achieve good performance or the best quality-speed tradeoff using the methods researched.