Low-complexity rate-distortion optimal transcoding of MPEG I-frames

Abstract

The transcoding of MPEG video streams is important in a variety of situations such as in rate reduction for small bandwidth downstream channels and in extracting a fast visual playback stream for digital video cassette recorders. This paper addresses transcoding of MPEG intra-coded frames by selectively copying DCT AC coefficients from the primary MPEG stream into the secondary MPEG stream. A rate-distortion optimal transcoding approach, as well as several computationally more attractive suboptimal approaches, is presented. The proposed solutions outperform the more traditional codeword extraction methods from literature.

Introduction

A typical MPEG encoder that compresses digital video for broadcasting purposes has a target bit-rate approximately between 4 and 10 Mbit/s. The choice of this rate is primarily determined by the available or allocated capacity of the terrestrial or satellite transmission link. In some cases these primary wide-band channels feed into secondary downstream channels of lower capacity. For instance, in a cable head-end a network operator may decide to allocate a smaller bandwidth to a specific video service than the bandwidth the service occupies on reception. Clearly this is at the advantage of providing more different services on the cable network however at the cost of a lower quality per channel. Another example is the provision of fast visual playback facilities in digital video cassette recorders. Fast visual playback, i.e., a fast forward or reverse playback mode which displays video pictures, is particularly complicated for MPEG compressed digital video stored on a helical scan video cassette recorder [2], [4], [5]. One solution to fast visual playback is to store (and display on fast forward or reverse) a low bit-rate stream on the digital cassette in addition to the original MPEG bit stream. This requires the real-time extraction of a low bit-rate MPEG stream when recording the primary MPEG video bit stream.

The problem of converting a compressed video stream such as MPEG into another (MPEG) compressed video stream of lower bit-rate is known as transcoding [7], [9]. Though with transcoding a wide class of format conversions for compressed video is meant, we confine ourselves in this paper to the efficient bit-rate reduction of the intra-coded frames (I-frames) of a given MPEG-2 video bit stream, without fully decoding and re-encoding (or re-quantization) the video signal. This last requirement is particularly important for cost-effective real-time systems. A typical configuration under consideration in this paper is the real-time conversion of a 10 Mbit/s primary MPEG-2 video stream into a 1 Mbit/s secondary I-frame-only MPEG-2 stream.

The solution strategy that we propose in this paper is that of selectively copying encoded non-zero AC coefficients or codewords from the primary stream into the secondary stream. Since no re-quantization nor inverse DCT operations are necessary, this method is inherently fast. Similar approaches have been suggested in [8], [10], [13]. The transcoding methods that we propose aim at rate-distortion optimality, and outperform earlier proposed techniques without increasing the complexity of the transcoder significantly.

In Section 2 various approaches to transcoding are discussed. Section 3 concentrates on rate-distortion optimal methods for coefficient or AC codeword selection, while Section 4 proposes a control strategy that governs the selection mechanism. In Section 5 an experimental evaluation is given for the transcoding methods described in this paper. The paper concludes with a discussion in Section 6.