Transrating of MPEG-2 coded video via requantization with optimal trellis-based DCT coefficients modification

Abstract

Requantization is one of the tools for bit-rate reduction of pre-encoded video to adapt it to various network bandwidth constraints. Several recent works propose using Lagrangian optimization to find the optimal quantization step for each coded macro-block, to meet a desired rate at minimum distortion. In this paper we propose to extend the Lagrangian optimization procedure by allowing the modification of quantized coefficients values, including setting their values to zero, in addition to quantization step-size selection. Coefficient value modification and quantization step-size selection are optimally done using a low complexity trellis-based procedure. The proposed requantization algorithm provides higher PSNR values than the Lagrangian-based optimization method that only handles the selection of quantization steps, and still does not exceed considerably its complexity.