Research On The High Efficiency Video Intra Coding

After several decades of development, the video coding technology is beingdeveloped in the direction of higher resolution, better coding performance, threedimension and scalability. Both the widely used video coding standard-H.264/AVCand the ongoing next generation coding standard-High Efficiency Video Coding(HEVC) choose the intra coding frame and the inter coding frame as their main kinds offrames. First, the intra coding frame only utilizes the spatial correlation of image, whichleads to a relative low compress ability, therefore, the performance of intra coding needsfurther improvement. Second, due to the various special applications of intra codingframe such as random access point and refresh synchronous frame, it is worthy to makeuse of the ability of intra coding to eliminate temporal error propagation and giveterminal users as good viewing experiences as possible. Besides, because the existingintra coding technology adopts different size of block partition types andmulti-directional intra predictions, the coding complexity is relative high when choosingthe final direction of the prediction. Meanwhile, the intra coding blocks are predicateddependent on the availability of the adjacent blocks in the former rows and columns, sothe parallel coding or decoding process cannot be realized among the intra codingblocks, which increases the difficulties for the implementation of real-time coding inhigh definition and ultra high definition applications. Therefore, how to reduce thecomplexity of the intra coding and increase the capacity of parallel computingeffectively has considerable value as well.Based on H.264/AVC, this paper firstly focuses on increasing the intra codingefficiency from the aspects of prediction, transform and quantization, and then proposesan intra refresh algorithm based on motion tracking to improve the ability of errorresilience. In addition, this paper proposed several algorithms on reducing thecomplexity of HEVC intra coding from the aspects of speeding up the encoder andimprovement of the parallel coding ability. The main research work and contributionsare listed as follows:1. In-depth analysis is conducted on the key technologies of intra coding. The researchstatus of intra coding is classified and summarized, and its existing problems aredetailed analyzed, which points out the main research fields. The overall frame ofintra coding system and its development tendency are studied comprehensively, and the intra coding technology and performance criteria are also analyzed in detail,which establishes research foundations for the later research.2. A new intra prediction coding unit is proposed. It improves the coding efficiency byboth prediction and transform. In the traditional intra prediction algorithms thatchoose square blocks as unit, some of the pixels in the bottom right of the codingblocks are far from the reference pixels and thus less correlation between them,which may lead to a relative high prediction error. To solve this problem, aline-based multi-directional intra prediction structure is proposed. It uses horizontalor vertical lines as its unit within the macroblocks, the pixels that belong to theadjacent rows and columns can be used to predict the current lines in severaldirections. The proposed structure can shorten the distance between the predictionpixels and the reference pixels, and thus, increase the prediction accuracy. Secondly,a One Dimensional Discrete Cosine Transform (1DDCT) is applied in this paper totransform the linear prediction residual, which has the property of directionaltexture distribution, and then the transform type is selected adaptively according tothe prediction mode, which can further increase the efficiency of the intra coding.Experimental results show that compared with the H.264/AVC, the proposedline-based intra coding framework reduces the bit rate by approximately4.2%at thesame Peak Signal to Noise Ratio (PSNR).3. An adaptive rounding offset for video coding is proposed. During the process ofquantization, rounding offset that controls the size of the dead zone is fixed, whichcan not satisfy the requirements of the video sequence that has divergent properties.For this problem, a dead zone control algorithm for quantizer that can adapt itselfaccording to the content of the video sequence is proposed. The rounding offset ischosen adaptively based on the minimum blocks’ discontinuities at the boundaries,which can help control the size of dead zone and decrease the quantizing distortion.The experimental results show that this algorithm increases both the objective andsubject quality of the reconstructed image after quantization.4. An effective intra refreshment algorithm is proposed. Intra refreshment ofmacroblocks is a kind of technology that increases the transmission robustness ofthe video sequence effectively. However, too many intra blocks may lead to adrastic increase of bit rate and also decrease the coding efficiency. To solve thisproblem, an adaptive intra refreshment algorithm based on motion tracking is proposed. For every frame between two successive intra refreshment frames, themotion affected regions are analyzed statistically to select the error sensitivemacroblocks and then intra code these macroblocks. To suppress the potentialspatial and temporal error propagation, we limit the reference number of interpredicted frame which is behind an intra refreshed frame. Experimental resultsshow that the proposed algorithm can obviously increases the objective andsubjective quality of reconstructed video image at various packet lost rates with lowcomplexity of computation.5. Based on the quadtree structure of HEVC coding unit, a fast intra coding algorithmis proposed. The scheme of HEVC intra coding extends the block based structure ofthe H.264/AVC, and provides more division patterns and prediction modes ofcoding units, which improves the coding performance, but at the same time itincreases the computational complexity of the encoder obviously. For this problem,based on the results of the smooth region detection, a fast intra coding algorithmthat can adaptively skip the split of the coding blocks that do not suit for its textureproperty is proposed. Experimental results show the proposed algorithm can speedup the encoder by about20%without considerable coding efficiency lost.6. An efficient parallel architecture for the emerging HEVC standard is investigated.In HEVC, the prediction process requires that the causal neighbors of the currentblock must be completely reconstructed before processing the current block. Itresults in a set of serial dependencies, which makes a parallelization of intraprediction to be a challenge, especially for the4×4intra prediction. In this paper, aparallel4×4intra prediction algorithm is proposed by limiting the reference pixelsof the4×4subblocks, which can reduce the dependencies on the neighboring pixels,at the meantime, the subblocks can utilize their different prediction directions.Experimental results show that it can achieve a satisfying intra parallelism withoutsignificant performance loss.