Video coding standards k. R. Rao, Do Nyeon Kim J. J. Hwang Springer 2014

P.5.250 See [E382] This paper explains clearly the fully pipelined architecture for intra prediction in HEVC and achieves a high throughput of 4 pels per clock cycle. It can decode 3840x2160 videos at 30 fps. In the conclusions the authors state “in the future work, we plan to implement the proposed architecture on ASIC platform to increase the system frequency, aiming at achieving real-time video decoding of higher resolution(higher than 4K videos)”. Explore this.

P.5.251 See P.5.250. The authors also state “Meanwhile the proposed architecture will be integrated with inter prediction engine, inverse transforming engine and other engines to construct the entire encoding/decoding system”. This is a major project. Implement this HEVC codec.

P.5.252 See [E281]. In the conclusions it is stated “Then, two flexible and HEVC compliant architectures, able to support the DCT of size 4, 8, 16, 32 have been proposed”. In beyond HEVC, DCT of size 64 also being considered. Extend these HEVC compliant architectures to support the DCT of size 64. See the sections BEYOND HEVC and PROJECTS ON BEYOND HEVC towards the end of this chapter.

P.5.253 Based on a set of Pareto-efficient encoding configurations identified through a rate-distortion-complexity analysis Correa et al [E383] have developed a scheme that accurately limits the HEVC encoding time below a predefined target for each GOP. The results also indicate negligible BD-rate loss at significant complexity reduction. Go through this paper in detail and confirm the results shown in Tables II – IV and Figs. 1 and 4-9 using the test sequences listed in Table I.

P.5.254 See [E384]. After describing the mosaic videos with arbitrary color filter arrays (Fig. 1), the authors propose a novel chroma subsampling strategy (4:2:0 format) for compressing mosaic videos in H.264/AVC and HEVC. They claim that this strategy has the best quality and bitrate compared with previous schemes.

For the seven typical RGB-CFA structures shown in Fig. 1 (captured mosaic videos) apply the demosaicking designed for each structure and as well the universal demosaicking and obtain the demosaicked full-color RGB images.

P.5. 255 See P.5.254. For the demosaicked videos convert from RGB to YUV and then to 4:2:0 six subsampling strategies (Fig.6). Convert from 4:2:0 to 4:4:4 YUV format and then to RGB reconstructed video. Compare these strategies in terms of CPSNR and CMSE for these demosaicked videos.

P.5.256 Implement the chroma subsampling strategy described in [E384] along with other strategies (Figs. 2-4) in both H.264/AVC and HEVC and confirm the results shown in Figs. 5-8 and Tables I-VI.

P.5.257 Apply the universal demosaicking algorithm based on statistical MAP estimation developed in [E387] to the seven typical RGB-CFA structures (see P.5.284) and obtain the demosaicked full color RGB images. Compare the effectiveness of this universal algorithm with those described in [E384] in terms of the PSNR. See the conclusions section in [E387].

P.5.258 See [E388]. Also the abstract. Implement the fast prediction mode decision in HEVC developed in [E388] and confirm the simulation results shown in Tables 1-3.

P.5.259 See P.5.258. Apply this technique to 8K test sequences and develop tables similar to Tables 1-3 described in [E388].

P.5.260 Review [E389] in detail. The last sentence in Conclusion section states, “The future work will focus on extending algorithm to the remaining coding structures (i.e., PUs and TUs) and other configurations in order to further expedite the encoding process with minimal impact on the coding efficiency”. Explore this and develop tables similar to Tables V-VII and draw the conclusions.

P.5.261 See [AVS12}. The authors proposed a fast intra coding platform for AVS2 (called iAVS2) leading to higher speeds and better balance between speed and compression efficiency loss especially for large size videos. The authors state “Owing to their (AVS2 and HEVC) similar frameworks, the proposed systematic solution and the fast algorithms can also be applied in HEVC intra coding design”. Go thru this paper in detail and explore how the speed up methods can be applied to intra HEVC using various test sequences based on the standard performance metrics.

P.5.262 See [E392]. SDCT and its integer approximation have been proposed and applied in image coding. The authors suggest the possibility to implement efficiently an integer SDCT in the HEVC standard.

Integrating inside HEVC may require a significant amount of work, as the transform has to be inserted in the rate-distortion optimization loop, and auxiliary information may have to be signaled. RD optimization is feasible but, the way at least the HEVC software is written, it is not necessarily easy. What can be easily done is to take the HEVC integer transform and rotate the transform using the technique developed in [E392] for obtaining a rotated transform. Implement this in all profiles and compare with the HEVC HM software. Consider both options.

P.5.263 See P.5.262 Using SDCT and its integer approximation in HEVC invariably results in increased implementation complexity. Investigate this thoroughly.

P.5.264 Performance comparison of SDCT with DCT is shown in Tables I, II and IV [E392]. However the block sizes are limited to 32x32. For super high resolution videos it is suggested that even larger block sizes such as 64x64 are suggested. Extend these Tables for 64x64 block size.

P5.265 See P.5.265. Extend these Tables to 4K and 8K video sequences.

P.5.266 Performance comparison of INTSDCT with DCT is shown in Table V [E392}. Extend this comparison to 4K and 8K video sequences and also larger block size such as 64x64.

P.5.267 In P.5.264 thru P.5.266 consider implementation complexity as another comparison metric.

P.5.268 Huang et al [E393] have developed false contour detection and removal (FCDR) method and applied it to HEVC and H.264 videos as post processing operation. They also state “It will be interesting to adopt it as part of the in loop decoding process (Fig.5.5). This idea demands further investigation and verification”. Investigate this in detail and see how FCDR can be embedded as an in loop operation besides the deblocking and SAO filters. Assuming FCDR embedding is successful, compare this with FCDR as a post processing operation in terms of PSNR, SSIM and subjective quality (See the corresponding figures and tables in [E393].) using various test sequences.

P.5.269 See P.5.268. Consider implementation complexity as another comparison metric in the FCDR process (in loop vs. post processing) in both HEVC and H.264.

P.5.270 Chen et al [394] proposed a novel block-composed background reference (BCBR) scheme and is implemented in HEVC. They claim that the new BCBR algorithm can achieve better performance than baseline HEVC. This technique is however limited to sequences captured by static cameras (surveillance and conference sequences). They also suggest, “for moving camera cases, the long-term temporal correlation due to background is also worthy of investigation and the block-composed sprite coding would be a good choice”. Investigate this in detail. Consider also encoding and decoding complexity as another performance metric.

P.5.271 See P.5.270 The authors in [394] conclude “We would like to extend our BCBR to more generic cases in our future work”. Explore this.

P.5.272 Min, Xu and Cheung [E397] proposed a fully pipeline architecture, which achieves higher throughput, smaller area and less memory, for intra prediction of HEVC. They conclude

“In the future work, we plan to implement the proposed

architecture on ASIC platform to increase the system frequency, aiming at achieving real-time video decoding of higher resolution.”. Implement this.

P.5.273 See P.5.272 Min, Xu and Cheung [E397] further state “Meanwhile, the proposed architecture will be integrated with inter prediction engine, inverse transforming engine, and the other engines to construct the entire full scale encoding/decoding system.’ Explore this in detail and implement the same.