Deinterlacing Algorithm using Edge Direction from Analysis of the DCT Coefficient Distribution

Abstract

We propose a single-field deinterlacing algorithm with compressed domain edge direction information (CD-EDI). This algorithm was verified using the DCT coefficient distributions and pixel domain edge direction information (PD-EDI) in conjunction with an analysis of the neighboring pixels. The CD-EDI, which was calculated in the DCT domain, was first addressed, so that we could categorize the edge direction. We studied the distribution of coefficients in a DCT-encoded the distribution of DCT coefficients in a DCT-encoded block by taking into account the orientations of the four edge directions: 0 (horizontal), pi/4 (increase diagonal), pi/2 (vertical), and 3 pi/4 (decrease diagonal). In addition, we studied the PD-EDI, which we acquired by identifying the small pixel variations at the six edge directions. On the basis of an edge-based line average algorithm, a PD-EDI was established within the operation window in order to lower the frequency of false judgments about the edge direction where a deinterlacing would he made. Finally, we propose an edge direction confidence (EDC) conception for each missing pixel, which is capable of revealing the accuracy of the edge detection. Since fuzzy sets can be represented by a membership function (MF), we used Gaussian MF to find EDC. The proposed algorithm has a simple EDC-evaluating structure of low complexity, which therefore makes it easy for implementation in hardware. In the extensive simulations conducted for different test sequences, the proposed W algorithm outperformed all of the other state-of-the-art deinterlacing methods.