Quantized DCT on H26L Test Model

This paper describes the implementation of a Quantized Discrete Transform algorithm (QDCT) for an H.26L test model. The QDCT is used for both MPEG and H.263 video coding [1]. The MPEG and H.263 coding are based on fractional 8x8 blocks. The new H.26L video standard uses a residue transform that is based on a 4x4 block and an integer transform. Several implementations of the QDCT are proposed in this paper to optimize the H.26L Codec. In the first implementation, the quantization is embedded in the DCT. In the second one, a multiplierless algorithm is proposed. In the third implementation, additional computational savings are achieved by adopting an early termination algorithm. Results are obtained using the QCIF (176 x 144) test sequence TELEPHONE with bit rate ranging from 41 kbps to 1.4 Mbps. I. Quantized DCT and its implementation on H26L test model 5.2: Usually the 4x4 quantized DCT transform coefficients can be written as follows: yk =yk / Q =  Σ cki xi  where cki = [ckcos ( 2i+1)kπ/8]/Q, Q is the quantization step, and   denotes the rounding operator. The corresponding matrix representations are y = Cx and y =  C x where y and y q are 4 x 1 column vectors, and C and C are 4 x 4 transform matrices. In a similar manner, we can write the necessary equations in the 2_D case, using the separability property, as follows: Y = CrXCc and Y =  CrXCc  Here, Cr and Cc are the corresponding matrices for the row and column transforms, and they are given by Cr = C/ Qr and Cc = C/ Qc