|M.Sc Student||Lavrentiev Michael|
|Subject||Transrating of Coded Video Signals via Optimized|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus David Malah|
Multimedia content provides rich information to consumers, but also poses challenging problems of management, delivery, access, and retrieval because of data volume and complexity. Digital video needs to be compressed for the purpose of efficient storage and transmission. The common solution is to produce or retrieve from storage a single high quality bitstream, and to match it to each end-user bandwidth constraints by transrating.
In this research work, we begin with a study of different MPEG-2 transrating approaches, based on requantization, and implemented them. Lagrangian optimization of the quantization step-size of each Macro-block is compared with a "simple" complexity-based transrating scheme; with cascaded decoding and encoding, and with the direct encoding of the original video sequence at the desired final bit-rate, using a standard TM5 encoder. To reduce requantization errors, MSE and MAP requantization decision approaches are added to the previously developed algorithms.
A novel extension of the Lagrangian optimization method, which optimally modifies quantized DCT indices, is developed and compared to existing methods. The proposed method outperforms all currently known requantization-based transrating approaches. To reduce the complexity of the proposed algorithm, we provide a low complexity trellis-based optimization scheme, and discuss other complexity reduction means as well.
Finally, we propose a simple approach for taking into consideration Human Visual System properties in the transrating process. The input pictures are segmented into areas of textures, smooth areas and boundaries, and the distortion of each block is weighted according to its type. This way the Lagrangian optimization allocates bits to the different segments according to their perceptual importance.
Perceptual observations, as well as measurements made by Tektronix's Perceptual Quality Assessment (PQA) tool, show an improvement of output video quality by the proposed schemes.