94 95TELE-audiovision International — The World‘s Leading Digital TV Industry Publication — 11-12/2014 — www.TELE-audiovision.com www.TELE-audiovision.com — 11-12/2014 — TELE-audiovision International — 全球发行量最大的数字电视杂志

• requires even less of the original video information• re-creates parts of the original video by artificial intelligence• saves on the amount of data required to transmit• makes use of software developed for gaming

AVS2 A New Video Coding Standard

FEATURE New Standard

HD video

96 TELE-audiovision International — The World‘s Leading Digital TV Industry Publication — 11-12/2014 — www.TELE-audiovision.com

FEATURE New Standard

AVS2 is a third generation cod-ing standard under development by the Audio Video Coding Stand-ard Workgroup of China. It will be a successor of AVS. The AVS standard is comparable in per-formance with the H.264/MPEG-4 AVC commonly used all over the world for coding HD video. The tests showed that both stand-ards achieved almost identical performance for HD signals. Per-formance is here understood as the PSNR (Peak Signal-to-Noise Ratio). Only for smaller resolu-tions (SD), MPEG-4 proved to be a little better. An important fea-ture of the AVS standard is its lower complexity of the encoder and decoder what makes it more practical.

Another reason for China developing the AVS standard is saving money by not paying royalties for MPEG-4. Most of the patents in AVS standard belong to Chinese companies and organiza-tions. They charge for them much less than the Western world companies do for MPEG-4. In this way, a Chinese consumer can save maybe 5-10% when buying a AVS receiver without MPEG-4 decoder.

But China didn’t stop at AVS and developed the standard further one. AVS2 will have a better compression ratio and thanks to that it will be more adequate for ultra high definition TV. Actually, AVS2 can be seen at the Chi-nese answer to the new HEVC/H.265 standard published recently by ISO/IEC and ITU.

In what ways will AVS2 ensure better performance? Well, the very accurate explanation is extremely complex, re-quires good background in mathemat-ics and only a narrow group of experts really can fully understand that. But what we can understand are at least

the basic concepts underlying the methods used in AVS2.

Let’s start with the texture analysis and synthesis. The readers more fa-miliar with computer games certainly understand that their software games synthesize various textures on differ-ent objects required in a game. And there is no need to store every pixel of the surface. The software program can create complex texture knowing only a small pattern of a bigger area. New compression algorithms in AVS2 can also do that. And instead of trans-ferring information of many pixels of a wavy sea or a distant flowerbed the AVS2 encoder will analyze what tex-ture is needed for this part of the pic-ture and will send to the receiver only this information (only a small picture). Now, the decoder in your receiver will fill in holes in an image by synthesizing non-repetitive parts of an image, as in inpainting.

Another interesting method is super-resolution based video coding. To put it simply, a high resolution image is reconstructed from multiple sequen-tial low resolution images. During this process high frequency modeling as well as spacial-temporal interpolation is performed. Interpolation means re-constructing correct values for an un-known image pixel located between known pixels. Located either in space (left/right/top/bottom) or time (previ-ous/next).

Learning based video coding is may-be even more interesting. The encoder analyzes the video sequence in which one or more objects are moving. It yields information about size, loca-tion and motion of the objects. Using computer graphics methods, it creates models of each object. It sends to the receiver decoder information about the model and animation information. This is sufficient for low resolution vid-

eo. To make it suitable for HD video, some additional information contain-ing residual pixel signal is sent. This is simply the difference between the model and the real image processed by the encoder.

Except for the above new concepts described above, there are also more improvements in the methods and algorithms used so far. So, AVS2 will take advantage of: Super-macroblock prediction, Adaptive Block-size Trans-form (ABT), Directional transform, Ad-vanced motion vector prediction and Rate Distortion Optimization Quantiza-tion (RDOQ).

We can say that up to the second generation (AVS and MPEG-4) video coding standards relied mainly on tak-ing advantage of various imperfec-tion of the human eye to achieve high compression ratio. AVS2 takes a step further. Some elements of the video will be in fact computer animations or computer generated textures resem-bling real things. This is another step away from transmitting the original video: what you see on your monitor is a brand new artificially created video, which looks as the original, but in real-ity it is re-created by using only some parts of the original video.

Compressed video already went away from a 1:1 transmission, as in the old analog times (the original is identical to the copy). Now even less of the original is needed, to re-created it.

Will AVS2 achieve similar compres-sion ratio improvement over AVS as HVEC has demonstrated over MPEG-4? Can it save up to 50% of the band-width? Some scientific papers report up to 37% of bandwidth reduction. This can change because the standard is not finalized yet and very few test results have been published. So let’s see where all this will eventually end.

How to artificially re-create a HD videoJacek Pawlowski