Improving rate distortion characteristics in offline video coding | |
| Author | Arachchi, Hemantha Kodikara |
| Call Number | AIT Diss no.TC-04-01 |
| Subject(s) | Video compression Digital video MPEG (Video coding standard) |
| Note | A dissertation submitted in partial fulfilment of the requirements for the degree of Doctor of Engineering, School of Engineering and Technology |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. TC-04-01 |
| Abstract | Digital video is robust enough to replicate and transmit almost perfectly and delicate enough to manipulate easily. This technology is being achieved a widespread recognition among the modern society. Powerful compression technologies, which facilitate the transmitting and storing high quality video materials at an acceptable cost, are behind this remarkable achievement. MPEG-1, MPEG-2, MPEG-4, H.261, H.263 and H.264 video coding standards were developed to serve for various digital video applications. MPEG-1/2 are the most popular compression schemes used in consumer video products such as Digital TV, VCD and DVD. Improving the quality of the coded video at a given data rate is the objective of this research. Two distinct approaches are taken in this regards, namely improving the rate control and improving the entropy coding efficiency. First, problems in TM5 rate control algorithm related to output buffer management are addressed. The significant problem in the above compression schemes is high fluctuation in output data rate over the video sequence. These encoders, in general, utilize a rate control algorithm in order to maintain the output data rate at a constant level regardless of the properties of the video sequence and differences in compression ratios of various picture types. Conventional TM5 rate control algorithm is not capable of managing the output buffer in the most efficient way to achieve the maximum possible average picture quality. Second, problems related to entropy coding efficiency in ZTE video coding algorithm are addressed. The wavelet block based scanning and the data organization technique is not capable of exploiting intrasubband redundancies among wavelet coefficients. In the first stage of the research, we analyzed the behavior of the TM5 algorithm at sudden and gradual scene transitions. At a sudden scene transition, almost all the macroblocks of the first anchor picture of the second scene are intra-coded when it is a predictive picture. Hence these macroblocks are coarsely quantized resulting in a poor picture quality. Furthermore, the bit allocation is not correct due to the use of complexity parameters derived based on the statistics of the earlier scene. To solve these two problems, we proposed an innovative algorithm to improve the video quality. The fixed encoding structure of the MPEG data stream is disturbed by extending or terminating the last GOP of the first scene such that the first anchor picture of the second scene becomes an I-coded picture. The complexity parameters of each picture types are, subsequently corrected with an iterative algorithm. Finally, P-pictures are provided with additional bits to cope with inefficient forward prediction during gradual scene transitions. Experimental results show that with the proposed algorithm, both objective and subjective quality of MPEG video can be improved significantly during scene transitions. Up to 2 dB objective quality improvement is demonstrated over sudden scene transitions. Further more, up to 0. 7 dB average gain in objective quality over gradual scene transitions is observed. The second stage of the research is to optimize the bit rate for offline encoding. The rate control problem is modeled as a constraint optimization problem. The channel rate, buffer occupancy and VBV delay constraints are considered in minimizing the rate dist01iion and the fluctuation of rate distortion in the proposed optimization problem. The less sensitivity of the human visual system for information losses in highly textured areas is also incorporated in deciding the final quantizer step size for each macroblock in a picture. We utilized evolutionary computation techniques to find the optimal set of quantization parameters for global rate control. The quantization parameters were allocated to a group of pictures (GOP) such that the average rate distortion is minimized under the constraint of the available bit rate. First, we used particle swarm optimization technique to solve this problem. The experimental results show that there is a significant improvement over the TM5 technique. Subsequently we used genetic algorithm and it shows a faster convergence than the particle swarm technique. These optimization techniques demonstrated an average PSNR improvement of about 4 dB. The final stage of the research is to develop a technique to improve the rate distortion characteristics of the ZTE video-coding algorithm. The wavelet tree concept proposed in the ZTE algorithm is utilized to eliminate insignificant wavelet coefficients. The remaining coefficients and symbols in each wavelet trees are subsequently reorganized according to the resolution level. These symbols and coefficients are scanned in a predetermined order over the subband. Resulting data streams are adaptive arithmetic coded. Proposed organization ensures that the entropy-coding model can exploit the intra-subband redundancies. Furthermore, aligning the scanning order of the wavelet coefficients with their native orientation maximizes the formation of runs of similar coefficients and thereby the entropy coding efficiency is maximized. This algorithm can save up to 50% bits in intra-coded pictures and up to 70% bits in predictive pictures. More than 3 dB gain of objective quality of both intra and predictive pictures is demonstrated. |
| Year | 2004 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. TC-04-01 |
| Type | Dissertation |
| School | School of Engineering and Technology (SET) |
| Department | Other Field of Studies (No Department) |
| Academic Program/FoS | Telecommunications (TC) |
| Chairperson(s) | Ahmed, Kazi M.; |
| Examination Committee(s) | Erke, Tapio J.; Teerapat Sa-nguankotchakorn; Afzulpurkar, Nitin;Furht, Barko ; |
| Scholarship Donor(s) | Asian Development Bank; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 2004 |