Research On Scalable Video Coding Oriented To Sample Video From Micro-beam Analytical Instruments

With the development of computer and network technology, the construction of scientific instrument network lab becomes possible. Through the Internet users can operate instruments remotely and view sample images with interaction of various means, such as text, voice, video, etc, as if users were personally on the scene. Compared with the traditional way of local experiment, the network lab can break through the limits of time, space and the number of people participating in the experiment, so that more researchers and students can use instrument. Network lab can promote sharing of instruments and improve utilization efficiency in some extent, so that it has important value of application.Micro-beam analytical instruments such as electron microscopes and electron probes, which are expensive and widely used, with high value of sharing, have become a research hotspot in the network lab.Sample image coding is one of the critical technologies about network lab for this kind of instruments. Sample video is defined as the dynamic sample image sequence formed in the process of searching for location of the sample and scanning the sample. Nowadays, sample images tend to be encoded in the way of video, but commonly, general video coding technology is adopted directly. Special coding algorithms based on characteristics of sample video are not deep enough, and there is large research space for improving their performance. In addition, MPEG-4, H.264 and other traditional single-layer video coding algorithms can only encode for different channel rates respectively, but they could not cope with the problems induced by the changeable network bandwidth and terminal diversity.Encoding video of highest spatial-temporal format for only one time, the latest scalable video coding standard H.264/SVC can get bit-stream with different frame rates, spatial resolutions and qualities, so that it can solve the problems of the terminal diversity and dynamic of network faced with in video applications. In this paper, H.264/SVC video coding standard was chosen to encode sample video. Due to its high complexity of encoding, this paper analyzed and used the characteristics of the sample video, studied fast algorithms for mode decision and motion estimation of high computational complexity, so as to improve the encoding speed. Meanwhile studying rate control algorithm for the coding structures with hierarchical B-frames contributed to improving the ability of controlling rate. All these work laid on the foundation for the establishment of network lab. The main contents are as follows:(1)The Fast Mode Decision AlgorithmThe multi-layer coding structure and excessive prediction mode of H.264/SVC lead to high computational complexity, which seriously influences the efficiency of real-time video coding. In order to increase the calculating speed of mode decision, a fast mode decision algorithm was proposed for spatial SVC. According to the distributed relationship of modes, the algorithm removed the prediction modes, which have less distribution and less influence on the quality of encoding. Meanwhile considering the computational complexity of each prediction mode, the rest prediction modes were divided into different mode sets. Based on the layer which is the current macroblock located at, and the rate-distortion cost from adjacent macroblocks in temporal and inter-layer directions, different thresholds of early termination were designed for each mode set, and a reasonable strategy of threshold was proposed to remove the unnecessary prediction modes. Experimental results show that, compared with the reference algorithm, proposed fast mode decision algorithm, which can be adapted to different bit rate environment, can save the encoding time by 78% with similar PSNR and bit rate.(2) The fast block matching algorithm based on sub-samplingFor reducing the computational complexity of motion estimation, a new fast block matching algorithm was proposed which based on sub-sampling. The algorithm divided the block matching criterion into three kinds:sum of absolute differences (SAD), half sub-sampling SAD, and quarter sub-sampling SAD, then analyzed their respective characteristics. According to the principle of Skip and BLSkip prediction mode, and the spatial correlation and the inter-layer correlation between macroblocks, the algorithm predicted the motion state of macroblock and chose a suitable block matching criterion combined with which layer the macroblock is located at. Besides, lossless early termination algorithm was designed to remove the calculation of redundancy in the process of block matching. Experiments show that compared with the reference algorithm, the proposed algorithm can approximately reduce coding time by 40% with similar PSNR and bit rate. The proposed block matching algorithm can be conveniently combined with mode decision and other types of motion estimation algorithm and so on, and it is suitable for sample video and nature video.(3)The fast motion estimation algorithm based on adaptive search rangeCommon motion estimation algorithm with a fixed search range generates a large number of invalid search points during encoding process. In view of this a fast motion estimation algorithm based on adaptive search range is proposed, in light of the motion principle of sample video. We divided macroblocks into different types using the similar method in previous algorithm, and utilized the characteristics that they have strong spatial and inter-layer correlation, to reduce the motion estimation search range by calculating the horizontal and vertical search range separately which based on motion vector difference in both reference frame and current frame. In addition, early termination algorithms were designed respectively for matching block searching and inter-frame prediction to further eliminate unnecessary search points. Experiments show that the proposed algorithm reduces by approximately 89% of the motion estimation search points, save about 62% of encoding time, while PSNR and bit rate maintain close to the reference algorithm.(4) The rate control algorithm for hierarchical B framesIn order to enhance the encoder's ability of rate control, a rate control algorithm for hierarchical B frame was proposed. The prediction of target bits for each P frame were improved by optimizing distribution of residual bits and limiting target bits for each P frame, and further improved the bit distribution of base units. According to spatial correlation between macroblocks, MAD linear prediction was further optimized. Moreover, this algorithm restricted QP range of P frame and B frame. The rate control algorithm for hierarchical B frame coding is not convenient for application, because it is lack of bit rate scalability between different temporal hierarchical levels. For this problem, a QP compute method of hierarchical B frame was designed. Experiments show that the proposed algorithm can realize accurate bit rate control, and flat bit-stream. In addition, it embodies the bit rate scalability of temporal scalable coding, so that bit-stream can adapt to different network environments of large viarety in bandwidth. The proposed algorithm is applicable to both sample video and nature video.