JVET-Y0104 AHG9: Transparency information SEI for transparent screens [E. Thomas, P. Andrivon, F. Le Léannec, M.-L. Champel (Xiaomi)]
It is asserted that a conventional video content displayed on a transparent display may result in a degraded experience for the viewer where important parts of the image may be lost such as logos, computer windows or characters, etc. Additionally, it is further asserted that the transparency effect perceived by the viewer of those transparent displays is controlled by the sample values of the displayed images and depends on the panel technology and the ambient lightning conditions.
Based on those observations, it is proposed to define metadata payload to signal the content creator intent with respect to how the decoded pictures should look like on a transparent display, i.e., which area should be transparent and to what degree.
To this end, a SEI signalling is presented which is asserted to address the above requirement and enable the transformation of the decoded pictures after decoding for both legacy and new content video content.
This was reviewed in session 18.
Values of mask (0..255) are encoded in the SEI message itself. Simple run-length coding can be used optionally. Mask could be sent for every picture.
It is suggested better using an auxiliary picture, and just signal its properties in an SEI message rather than signalling the mask values in the SEI message. With the proposed method, the bit rate for the mask might become higher than for the video. Such an SEI message should be simple to define, very similar to alpha channels.
How close are transparent displays in terms of introducing services specifically for those?
Generally this is an interesting domain, typical use case where an SEI message is beneficial. Suggesting an updated version is recommended.