P-Picture Based H.264 AVC to H.264 SVC Temporal Transcoding

This paper proposes a transcoding method from a standard (P-picture based) H264 advanced video coder (AVC) bitstream to a H264 scalable video coder (SVC) bitstream with temporal scalability. This method avoids full AVC decoding followed by full SVC re-encoding and is performed in two steps. The IPPP... AVC bit stream is first transcoded to an IpPpP AVC bitstream, where P denotes a reference image and p is a non reference one. Then, the IpPpP... AVC bitstream is transformed to an IpPpP SVC bitstream by syntax adaptation. The resulting SVC bitstream is thus fully compatible and can be decoded e.g., by the JSVM reference software tool. We present a theoretical analysis of the first step of the proposed transcoder with two results. First, we prove that the transcoding process does not introduce any error propagation. Second, we present an evaluation of the transcoding errors bounds compared to full AVC decoding followed by full AVC re-encoding process, denoted in the rest of the paper as the perfect transcoding. Simulation and experimental results are presented for "foreman" CIF format sequence encoded at a frame rate of 30 frames per second. Performances are discussed in terms of coding efficiency and algorithm complexity. Compared to IPPP AVC to IpPpP AVC perfect transcoding, the proposed transcoder presents a 47% complexity reduction with almost the same quality for the enhancement layer. The resulting bitstream thus benefits of the traditional properties of scalable videos. Moreover, the base layer obtained from the decoding of the AVC/SVC transcoded bitstream presents an equivalent performance to that obtained from full AVC-decoding followed by AVC re-encoding of the even frames.