Contents adaptive deinterlacing based on local content classification

In contents adaptive deinterlacing methods, accurate contents classification is important to minimize deinterlacing artifacts. The adaptive dynamic range coding (ADRC) method is widely used for local video contents classification because it has low complexity. However, since the ADRC method coarsely classifies local video contents, its performance is rather limited. For accurate local video contents classification, we propose a modified ADRC (MADRC) method. While the ADRC method encodes each pixel using 1-bit, the proposed method encodes each pixel using 2-bits by dividing into more detailed intervals. Encoded bits are concatenated together to form a class. We compute vertical-temporal (VT) filters using the least square solution for each class classified by the MADRC method. These VT filters are obtained from progressive videos in advance. Then, we adaptively apply these VT filters to interlaced video based on the local video contents classification results. To evaluate the proposed method, we conducted experiments using 13 CIF progressive video sequences. Experimental results show that the proposed deinterlacing method showed 1-3 dB improvement in terms of PSNR compared to existing methods.