We’ve been recently approached by a customer who intended to monitor the appearance of specific video ads in a video stream. Since a video stream can be simply decomposed to series of frames (pictures), it appears that our technology can be also used in such cases.
Recognize.im image recognition solution identifies items in a photo on the basis of matching them against a constructed database of source images. A database of source images is created as a part of each implementation and after a necessary pre-processing (indexing) allows to determine whether the selected image contains one or more images from the database. As a part of indexing the database, every photo is processed and examined for distinctive features. Based on this, it is possible to create easy to search index of photos.
As the customer had a number of video ads, we managed to separate distinct frames out of them and built a searchable index. We can get plenty of frames even for relatively short video advertisement but we usually pick only a few of them (eg. 10). After that we are ready to monitor the video stream and look for those frames. Even if there is a certain number of variations of ads – we can compose their “fingerprint” by choosing specific frames as the reference images.
Our technology operates in a pattern-matching way, therefore this case suits it perfectly – we have patterns extracted from original ads, we then divide the on-the-fly video stream into separate frames and see if there’s a match. Since our technology is very fast in performing such matching (200-500 milliseconds per frame) it’s possible to get near-real-time analysis.
Such cases can be then used for media monitoring purposes like automatic checking of video ads exposures.
Contact us if you are interested in using image recognition for video fingerprinting.
Recognize.im Business Development Director