Abstract
This paper describes a macroblock-based optimization technique for quantization parameter selection that reduces re-quantization errors in MPEG transcoding. Having found that the re-quantization with a larger quantization parameter but smaller than the double of the original one should be avoided, we start at choosing the better re-quantization parameter from two alternatives; the original quantization parameter or its double. Furthermore, we investigate a rate distortion model of residual errors in motion compensation over frames. We extend the optimization scheme to enable us to choose the re-quantization parameters from a wider range according to the rate distortion model. The proposed scheme outperforms a non-optimized typical transcoding scheme with 0.5-1dB improvements in PSNR.
