Research On Algorithm Of Fast Mode Decision In Scalable Video Coding

To achieve the requirements of diverse bitstream video end-users within the rapid expansion of digital video coding, a lot of advanced tools have been widely applied into the video codec. The ITU-T and MPEG form the Joint Video Team (JVT) to develop the scalable video coding (SVC) as the scalable amendment of the H.264/AVC standard. Being an extension of the previous video coding, SVC enables one single bitstream to support temporal, spatial, quality scalabilities and their combinations. However, these superior methods and technologies introduced in SVC also generated the enormous calculated amount while providing various services.In order to decrease the computational complexity, this thesis proposes a series of fast solutions to reduce the encoding time of SVC exploiting the relationship of macroblock modes in the features of the scalable prediction structure at different applications. Furthermore, these fast algorithms are easily forged into the special encoding hardware.First, a fast prediction mode decision algorithm is proposed for the prediction structure of the full-intra pictures under the condition of high intra scalable profile level, which is suitable for high definition television or other high fidelity devices. This algorithm uses the prediction modes of encoded macroblocks in base layer to decrease the candidate modes of the current enhancement encoding macroblock. The simulation results show that the algorithm reduces nearly one third of the encoding time on average with the neglect loss in quality compared with the standard fast prediction algorithm submitted by JVT.Second, two different fast mode decision algorithms have been proposed for spatiotemporal combination scalability. The former algorithm named region best-mode fast algorithm, which is suitable for the situation of less spatial layers, utilizes the neighbouring region macroblock modes of the co-located reference modes to cut down the candidate modes for the current enhancement encoding macroblock. This algorithm chooses different ranges of the neighbouring macroblocks according to the different temporal levels in temporal scalability of each encoding layer. That range becomes bigger when the temporal distance is larger. And the statistics of these neighbouring macroblocks with different temporal ranges form a candidate macroblock mode group for the current encoding macroblock. The latter algorithm uses the candidate modes of the co-located reference macroblocks including their best and second-best modes to determine an encoding mode for the current enhancement macroblock. This algorithm sorts those best and second-best modes as a candidate list. Those modes which have the lower rate distortion costs in this mode list will be taken grant as the candidates of the current encoding macroblock so that the current macroblock only needs little search computation for finding a potential best mode. With the simulation results, these two algorithms can reduce above 30%encoding time on average without any big loss in encoding quality.Last, this thesis proposes a novel efficient and low complexity prediction structure for CGS and provides a corresponding fast mode decision algorithm under this condition. This algorithm designates different weight coefficients for the co-located and its neighbouring macroblocks in base layer in order to deduce a reasonable best mode for the current enhancement encoding macroblock. And this algorithm can directly calculate the current best mode ignoring the temporal reference information in enhancement layer. By applying the adaptive macroblock predication method, the experimental results show that the algorithm can approach the optimal encoding outcomes which can achieve the best encoding results using SVC original algorithm.