NEC develops image identification technology that can instantaneously detect illegal copy movies even with accompanying data modification

It seems that NEC has developed a video identification technology that can instantaneously detect copied videos illegally uploaded on the Internet.

YouTubeWith the spread of video sharing services such as video sharing services, it has become possible to publish movies widely without having to be professional in video, but recently it is not my work but selfish Illegally copied videos uploaded to. Measures include visual checking, embedding digital watermarks in the content, etc. However, it is difficult to deal with when the amount is enormous, and it was difficult to discover if it was short cut or altered.

However, with the technology developed this time, it is possible to detect at high speed and high precision not only with copying but also with data alteration.

Details are as below.
Developed image identification technology that can instantly detect illegal copy video (May 7, 2010): Press Release | NEC

In this technique, fingerprint information (video signature) for identifying the video from the original video is generated and compared with other video. It seems that it is possible to detect not only copy videos but also high speed and high precision even if you shoot original with camera, insert telop, or copy it analog. By doing this, if content right holders and distribution service providers register original images in advance, illegal copy movies are automatically detected and it becomes possible to block illegal uploads in advance.

Regarding the signature, it seems that a plurality of areas of various sizes and shapes are set in pairs at various positions of each frame (frame) of the image and the difference in brightness in each pair area is extracted Since it is different for each image like a fingerprint, it can be detected even for animations that have been subjected to analog capture, re-encoding, telop insertion or the like.

In addition, when the difference in brightness in the region is small, the reliability is lowered, and by using importance on highly reliable information and using it for identification, highly accurate identification rate and low false detection power can be obtained It is realized. Since the signature has high discrimination performance, it seems to be able to correctly detect even a short scene of about 2 seconds which could not be detected in the past.

The size of the signature is 76 bytes per frame, and the memory capacity to hold the movie to be checked is small. Therefore, it is possible to check for about 1000 hours in one second even with the processing power of the household personal computer.

This technology is ISO / IEC 15938-3 / Amd.4 final at the 92nd ISO / IEC JTC 1 / SC 29 / WG 11 (commonly called MPEG) conference held in Dresden, Germany from 19th to 23rd April It is approved as a draft standard and will be issued as an official standard plan around September after ITTF member countries vote in the future. Also, it will be held at Tokyo Big Sight from 12th to 14th MayThe 13th Embedded System Development Technology ExhibitionIt will be exhibited at.

On YouTube, a warning is issued to users uploading videos matched with original videos by video identification technology, but can this be the same with other services as well?