Frequency domain down-conversion of HDTV using an optimal motion compensation scheme

In the MPEG-2 Test Model 5, the down-conversion of an interlaced sequence is obtained by preeltering and sub-sampling each eld of the image sequence after it has been fully decoded. Although the quality is very good, the cost of such a system is quite high due to large memory requirements. As a result, low-resolution decoders have been proposed to reduce some of the costs incurred by this scheme. Here, incoming DCT blocks are subject to a down-conversion process within the decoding loop, hence the motion compensation is performed using the down-converted images. In past work, it has been proven that the optimal lters for performing this motion compensation are intimately related to the method of down-conversion. Therefore, the choice of down-conversion lter is viewed as the primary variable aaecting the quality of the down-converted sequence when such an optimal motion compensation scheme is considered. In the conventional method of frequency domain down-conversion, the 4x4 low frequency coeecients are extracted from each 8x8 block. Two problems arise from this method: rst, the discarding of high frequency data will introduce a disturbing amount of drift, and second, severe blocking artifacts will result in areas of large motion. To remedy the drift problem, a new method of down-conversion which better preserves high frequency data is presented { this method is referred to as frame-based frequency synthesis. Then, to overcome the blocking artifacts, the previous method is extended to a eld-based frequency synthesis method. Our simulation results clearly indicate that the amount of drift can be signiicantly reduced by retaining high frequency data and that severe blocking artifacts are eliminated by using the eld-based method. Additionally, the quality achieved by the proposed eld-based frequency synthesis is much closer to the high quality results produced after full-resolution decoding.