Direct 3-D DCT-to-DCT resizing algorithm for video coding

This paper proposes a direct DCT-to-DCT resizing algorithm for 3-D DCT based video coding. An 8 times 8 times 8 cube is resized to three modes along temporal dimension: a single 8 times 8 block, a downsized 8 times 8 times 4 cube and two 8 times 8 times 4 cubes. The mode selection is based on the local motion activity and determined after 2-D DCT on each block. In addition, the proposed resizing scheme directly resizes the DCT blocks in the transform domain using the DCT-to-DCT algorithm. Compared to traditional resizing approaches the proposed algorithm does not require the inverse transform and the computations in the spatial domain, thus is superior to other methods in terms of complexity. The proposed model is evaluated with the baseline video codec and the reference codec in the literature. Experimental results show a promising performance in terms of both coding efficiency and computational complexity. Potential applications could be for portable digital devices with low-power processors and other areas with real-time requirement.