|M.Sc Student||Leventer Amir|
|Subject||On Bit-Rate Control in Video Coding|
|Department||Department of Electrical Engineering||Supervisor||Professor Moshe Porat|
Many video services use pre-encoded video for the distribution of video programs to end-users. The transmission of compressed video over channels with different capacities may require a reduction in bit rate if the transmission media has a lower capacity than the capacity required by the video bitstream, or when the communication network is congested. This thesis addresses the specific bit rate reduction problem of previously compressed MPEG-2 video.
In recent years several techniques for bit rate reduction of compressed video have been introduced. Although these techniques can be used in some applications, they cannot be regarded as generic video transcoding methods due to either their high computational complexity or major drift errors in the transcoded pictures.
A naive solution for the bit rate reduction of compressed video problem is a cascade of decoder and encoder. This solution is very expensive in computational complexity. The search of motion vectors in an encoder takes about 80% of its computational complexity. Moreover, this solution suffers from degradation of the output video quality due to imprecise decoding while computing the IDCT/DCT transformations and saturation of coefficients. Recently, new methods for bit rate reduction were developed which are more efficient in complexity and are based on information resides in the coded input video bit stream.
In this work, we propose a simplified and optimized transcoder based on the special properties of the Lagrangian multiplier optimal algorithm. Our algorithm enables to select the optimal requantization factor based on specific quantizer scale ratios on the Macroblock (MB) level that lead to minimal transcoding errors while creating a low bit budget.
Moreover, we suggest a new bit allocation technique that takes into consideration the drift errors in Open Loop transcoders. This technique enables to minimize the drift errors in open loop drifted architectures. Our global (frame) bit allocation algorithm could be used as guidelines for bit allocation when implementing Open Loop Bit Rate Reduction (BRR) systems, just as Test Model 5 (TM5) is used for the encoding process.
The experimental results show that our algorithm, based on these two methods, improves the quality of the output video for both motion compensation closed loop and for open loop architectures, while reducing dramatically the computational complexity of the algorithms.