Optimization Method For Stereoscopic Video Coding

As the development of the techniques of video signal processing, displaying and communication, the coding method of stereoscopic video become more and more mature, and the application of 3D video become more and more wide. The study subject of this thesis is the optimization of the coding and transmision of 3D video.First, a stereo video coding method considering the human vision property is proposed to reduce the redundancy of the human vision. When watching 3D video there is a visual masking effect. That is, the total quality of the stereo video is determined by the channel which has the higher quality than the other one if the quality difference between the two channels is not larger than a threshold. By maintaining the quality difference around the target value fewer bits are generated to coding the stereo sequence while equivilent subject quality is achieved. That is to say, higher coding performance is achieved. To maintain the quality difference the relationship between the quality differnce and the left and right quantization parameters is modeled. This model can then be used to calculate the quantization parameters used in coding to make the quality difference to reach the target value. This method can reduce the visual redandancy to a great extent, thus reaching higher perfomance.Second, in order to control the output bitrate under the limit of the bandwidth a stereo video rate control method is proposed. In the algorithm the bit allocation is conducted for the GOP layer, the layer of the picture pair in the GOP and picture layer in the picture pair. Then the quadric model is used to calculate the quantization parameter. The experiment results show that the control error between the actual bitrate and the target bitrate is below 2 percent while the relative quality between the left and right channel is stable. And more over, higher coding efficiency is achieved.At last, a new rate-quantization model called Exponential Model which is more stable and accurate than traditional quadric model is proposed to control bitrate more accurately and reduce the complexity in rate control. In the algorith the model parameter is not updated, thus reducing the computing complexcity. The experiment results show that a relative accurate control is achived at high bitrate, and higher coding performance is achived because of the use of the human visual redundancy.