A Fast Inverse Motion Compensation Algorithm for DCT-Domain Video Transcoder

Abstract

The existing methods to perform motion compensation (MC) in discrete cosine transform (DCT) treat each 8 times 8 block as a fundamental unit and, therefore, involve the high cost of reconstructing prediction frames, especially when the half-pixel motion vectors (MVs) are involved. The proposed method operates on a block of variable size 16 Ny times 16 Nx, where Nx and Ny are the number of adjoining 16 times 16 macroblocks with common MV, along the horizontal and vertical directions, respectively. We demonstrate that a basic operation in reconstructing an 16 Ny times 16 Nx DCT-MC block, both with integer and half-pixel MV, can be represented as a multiplication by fixed matrices, and the computations can be greatly simplified through decomposition of DCT/ IDCT operations. Experimental results using cascaded DCT domain transcoder show substantial reduction in computations at quality close to pixel domain transcoding.