trace.moe (source code) is a web service that can identify an anime from a single frame/screenshot. After loading a screenshot into the service, it returns the anime name, the episode number, and the timestamp of the frame in the episode (accurate to a hundredth of a second).
My idea is, maybe we could have something like that for broadcast and cable television, to identify the currently-playing TV show or movie. EPG information is not always available or correct, so it would be useful to be able to identify the video solely based on the video data, and then to overlay that information on top of the video.
Unfortunately, TV signals are usually DRM-encumbered, and even after the initial decryption and decoding by set-top boxes, the video data is usually re-encrypted with HDCP when sent to the TV set. Furthermore, it is almost impossible to legally obtain most television shows and movies as DRM-free video files, since all the legal methods of viewing that content (e.g., streaming services, Blu-ray/DVD) are DRM-encumbered. These two issues combined make this idea a nonstarter, legally, since all the broadcast/cable TV video we would want to identify would be encrypted, and we wouldn't be permitted to decrypt the video files needed to generate the video frame index database.
trace.moe only exists because they were able to download a large quantity of anime illegally--without the restrictions of DRM--in order to generate their video index, and it only continues to exist because no one has sued them or sent them (or their VPS host/ISP) a DMCA takedown notice yet.
If we had the right to remove DRM from TV signals or the output of set-top boxes, we would be able to legally access the raw video data and send screenshots of it to a computer, and then we could use either a local or cloud service to perform the identification of the TV program. And more importantly, we would be allowed to strip the DRM from the legally-purchased streaming video and Blu-rays/DVDs, which we'd need the raw video from in order to generate the indexed corpus of video data necessary to perform scene and program identification.
trace.moe (source code) is a web service that can identify an anime from a single frame/screenshot. After loading a screenshot into the service, it returns the anime name, the episode number, and the timestamp of the frame in the episode (accurate to a hundredth of a second).
My idea is, maybe we could have something like that for broadcast and cable television, to identify the currently-playing TV show or movie. EPG information is not always available or correct, so it would be useful to be able to identify the video solely based on the video data, and then to overlay that information on top of the video.
Unfortunately, TV signals are usually DRM-encumbered, and even after the initial decryption and decoding by set-top boxes, the video data is usually re-encrypted with HDCP when sent to the TV set. Furthermore, it is almost impossible to legally obtain most television shows and movies as DRM-free video files, since all the legal methods of viewing that content (e.g., streaming services, Blu-ray/DVD) are DRM-encumbered. These two issues combined make this idea a nonstarter, legally, since all the broadcast/cable TV video we would want to identify would be encrypted, and we wouldn't be permitted to decrypt the video files needed to generate the video frame index database.
trace.moe only exists because they were able to download a large quantity of anime illegally--without the restrictions of DRM--in order to generate their video index, and it only continues to exist because no one has sued them or sent them (or their VPS host/ISP) a DMCA takedown notice yet.
If we had the right to remove DRM from TV signals or the output of set-top boxes, we would be able to legally access the raw video data and send screenshots of it to a computer, and then we could use either a local or cloud service to perform the identification of the TV program. And more importantly, we would be allowed to strip the DRM from the legally-purchased streaming video and Blu-rays/DVDs, which we'd need the raw video from in order to generate the indexed corpus of video data necessary to perform scene and program identification.