Techniques and algorithms for video transcoding

Abstract

Video transcoding becomes an important role for a video server to provide quality support services to heterogeneous clients or transmission channels. Video transcoding can be viewed as a process of converting a previously compressed video bitstream into a bit stream of different nature or lower bitrate bitstream. From this point of view, the focus introduced in this thesis is a strategy to convert the previously encoded video into other encoded format or reducing the bitrate of the encoded video according to the bandwidth requirement. To achieve this purpose, there are different ways to perform video transcoding. In this thesis, we will first look at the conventional video transcoder using three different techniques: frame skipping, video downscaling and requantization of DCT coefficients. The most straightforward method to implement these approaches is to cascade a decoder and an encoder, commonly known as pixel-domain transcoding. The input video bitstream is decoded in the pixel domain, and the decoded video signal is re-encoded at the desired output bit rate using frame skipping, requantization or video downscaling technique according to the capability of the clients' devices and the available bandwidth of the network. The major problem in this approach is that high processing complexity, large memory size, long delay and re-encoding errors will be introduced. In the first half of the thesis, our studies focus on the homogeneous video transcoding. A DCT-based video frame skipping transcoder is proposed to provide low computational complexity. Using the proposed direct addition of DCT coefficients technique, the DCT coefficients can be re-estimated directly in the DCT domain without performing full decoding and re-encoding process. For the motion compensated macroblock video transcoding, a DCT coefficient re-composition algorithm is proposed to reconstruct the re-estimated DCT coefficients from the pixel domain data and the transform domain data. For video downscaling, a new architecture to obtain resampled DCT coefficients in the DCT domain by using the split and merge technique is proposed. In this approach, the macroblock is splitted into the dominant region and the boundary region. Then, the dominant region of the macroblock can be transcoded in the DCT domain using the proposed transcoding operators to achieve low computational complexity. By transcoding the boundary region adaptively, the computational complexity can be reduced and the re-encoding error introduced in the boundary region can be controlled more dynamically. In the second half of the thesis, we address problems of heterogeneous video transcoding which is to transcode the video from one format to another format. A new compressed-domain heterogeneous video transcoder is proposed to convert a B-picture into a P-picture by making use of the techniques of motion compensation in the DCT domain and indirect addition of DCT coefficients. A set of equations, the formulation of the problem and solution are given in this new research area. Since the B-picture is transcoded to the P-picture, this new reconstructed P-frame will be used as a reference frame for the next incoming frame. Therefore, the next incoming P-frame needs to be transcoded in order to reference to the transcoded P-picture. A fast transcoding using the backward subtraction of the DCT coefficients is proposed to transcode the P frame in the DCT domain to achieve low computational complexity. The H.264 video coding standard has been filed recently. This creates an important need for transcoding technologies that transcode the widely available H.263 compressed videos to H.264 compressed format and vice versa. Direct conversion of DCT coefficients to Integer Transform(IT) coefficients in the compressed domain is proposed for the H.263 to H.264 video transcoder. A set of operators is derived for converting the DCT coefficients to integer transform coefficients in the compressed domain directly. Also, a set of operators is proposed to obtain the motion-compensated integer transform coefficients(MC-I) form the incoming DCT coefficients directly for inter frame vide conversion. An approximation form of the DCT coefficients is proposed to speed up the transcoding process for low bitrate applications. In addition, the integer approximation of the operators is proposed to avoid the floating point implantation for the proposed transcoder. It is exciting to report in this thesis that significant gains in terms of computation and high quality of video can be achieved by employing our proposed approaches. Undoubtedly, these advance techniques can enable the video transcoding to become more efficient and provide good quality video in practical situations.