An effective hybrid video deinterlacing algorithm

HDTV system networks are expanding, but most TV systems still use traditional interlaced scans. In order to display the traditional TV signals on HDTV, deinterlacing and scan up conversion are applied. Even if all traditional TV systems were replaced with HDTV system, deinterlacing would be required to convert the video sequences that were originally recorded in interlaced scan to a progressive scan. Deinterlacing can improve the video quality by increasing the vertical resolution of the picture and by removing interlacing artifacts. Nowadays, the greatest challenge for deinterlacing is balancing the trade-off between implementation complexity and a reasonable image quality.;In this thesis we propose an effective hybrid deinterlacing algorithm in which a motion compensated algorithm is combined with an edge-based method based on the reliability of motion vectors. A five-field motion compensated algorithm that works by using the amount of vertical motion within the reference fields is proposed to achieve the maximum vertical resolution improvement.;A new measure for motion vector reliability assessment called reverse motion estimation (RME) is also proposed to qualify the motion vectors. The method tries reversing the calculated motion vector to confirm its validity. This measure was used in the hybrid algorithm to recognize unreliable motion vectors and switch to an edge-based method to prevent artifacts. The RME measure has a very low computational complexity compared to the other measures while being more efficient.;A Pattern-Based Directional Interpolation (PBDI) deinterlacing algorithm is introduced. The algorithm can be used in the hybrid method in the case that unreliable motion vectors are encountered. It compares three-pixel patterns in the upper and lower lines to detect twenty one edge directions and to provide clear and smooth edges with less probability chance of misleading edges. Pattern comparison is done by a combination of the sum of absolute differences and a gradient-based difference.