Xiph.Org explains the technical specifications of the next generation codec's favorite "AV 1"

Next-generation video codec developed by Alliance for Open Media (AOM) aiming for license-free "AV1"Butofficial releaseit was done. Joining AOM and involved in the development of AV1Xiph.OrgI am talking about high compression technology incorporated in AV1. The first part of the series over several times is described about "Chroma from Luma prediction (CfL)".

next generation video: Introducing AV 1
https://people.xiph.org/~xiphmont/demo/av1/demo1.shtml

Next generation codec of free "Daala"Xiph.Org joins the AOM and is cooperating in the development of AV1. At first, it was Google's "VP 9", Cisco's "Thor" and Xiph.Org "Daala" who celebrated the codec which is the basis of the next-generation codec AV 1, but the risk of conflict with existing patents The lowest VP 9 was elected. According to Xiph.Org, at that time, Daala said that the wrapping approach and the clock domain technology were not matured and were not chosen as the base codec, and the decision itself that VP 9 was chosen was reasonable He seems to think.

Although it is not selected as a base codec, Daala, many of the technologies adopted by the same codec are also incorporated in AV1. Among the technical specifications of AV1 trying to settle, Xiph.Org is going to explain the technology adopted in AV1 in multiple times with the official blog, the first time focusing on the technology "CfL" It is explained.

◆ CfL
In AV1, saturation determination method based on luminance prediction "Chroma from Luma prediction: (CfL)"Has been adopted. CfL is a technique for predicting color from the saturation of an image based on luminance value (Luma) as its name. When the luminance value is first decoded, CfL predicts the color. If this prediction is good, you can reduce the data size by reducing many of the coded color information.

CfL itself is not a new technology, it is also adopted by other codecs. The difference between CfL of LG, LM, Thor, Daala, HEVC (H.265), AV1 is as follows. Xiph.Org says that the final AV1 CfL is based on the implementation of Daala, preventing the complexity of the decoder and encoder from increasing and improving the compatibility and accuracy of the encoded model.


Improvement of intraframe prediction performance
According to Xiph.Org, compression technology is "prediction technology" if it keeps on. The accuracy of predicting Inter-frame (inter: frame) between frames is always pursued in the development of new codecs. However, as the prediction accuracy of the intermediate frame increases and the size decreases, the size difference from the key frame (I frame) that is difficult to compress is increasing.

The following image shows the bit rate of movie 60 frames. The leftmost keyframe is 20 to 30 times the size of the intermediate frame.


As the size of the intermediate frame decreases, the percentage of key frames occupying the entire movie size is increasing, and the need to reduce the size of the key frame is emerging. Therefore, intra-prediction is used to reduce keyframe size. And Xiph.Org says that the importance of intra-frame prediction technology improvement is increasing because it also leads to improved accuracy of CfL.

Besides this, since there are written about various technologies adopted in AV1 as a theme of CfL, there should be something that can be gotten by people with technical interest in codecs.