How Decoding the Genome Helps Compare Video Clips to Locate Duplicated Films

The search engine developed by Dr. Alex Bronstein from the Faculty of Engineering at Tel Aviv University, allows you to take a clip from a movie and search for copies of it among tens of millions of movies, at the same speed as the textual search is carried out using a method similar to searching a genetic database. .

Dr. Alex Bronstein
Dr. Alex Bronstein

Imagine a search engine into which you enter a movie, and get back as search results movies in which that segment appears. This is one of the applications possible thanks to an algorithm developed by Dr. Alex Bronstein from the Faculty of Engineering at Tel Aviv University. This application will be the bread and butter of movie producers fighting pirated copies.

The system is so accurate that it is able to distinguish sections of the film even if they have been edited or processed, including a change in resolution. It is not only about locating identical copies of the film, but also providing information to the searcher from which second to which second in different clips clips from the film appear.

According to Dr. Bronstein, today copying videos and transferring them on BitTorrent or other file sharing programs is commonplace, and the movie studios don't like it, to say the least. This way it will be possible to know who used clips from the movies we own, and even the location of the clips within the movies After the knowledge reaches the owners of the rights, they can decide whether to sue the pirates. "We took advantage of the analogies between genetic series and series of video scenes. We developed a system that condenses each movie into a unique series of characters that represents it similar to a genetic series in which the bits represent DNA segments. The method actually allows searching for a video segment with the same efficiency as bioinformatics practitioners use to locate a sequence of genetic bases.

Dr. Bronstein and his colleagues were able to use the method to build a video search engine that, when fed into the input, the engine finds within a short time copies of it among hundreds of thousands of video clips, literally a needle in a haystack.

Today, the search for video clips (for example on YouTube) is done simply by searching the text accompanying the movie, and it turns out that there are those who abuse this, when they attach misleading text to the movie, which causes wrong search results.

However, the researchers were not satisfied with only the complex possibility of identifying a segment in a sea of ​​videos as it is, but also used computer learning techniques in order to recognize the film even if it had been edited (just as bioinformatics systems know how to identify genes that have undergone a mutation. "We filmed a segment with a cell phone camera from a TV screen, and then we uploaded it to the network, and let the system search, it recognized both the original copies in good quality and the improvised copy, which by all accounts are not 100 percent identical, and not even close to it."

According to Dr. Bronstein, a comparison of the search engine he developed with the similar engines that exist today shows that the new algorithm makes it possible to perform a faster search, when a much smaller amount of data is kept in the database compared to other systems that perform the same activity.

To be more optimistic, one can think of a situation where the rights holder finds copies of his films on a file exchange site, but instead of blocking them or deleting them, he simply plants advertisements in the pirated copies and enjoys their profits.

Another use is to link the visual world with the textual world. When we watch, for example, a DVD movie, the system will be able to identify where and which movie we are watching and download a translation file from the web, or integrate into it targeted advertisements adapted to the movie's content. Or make the movie interactive - clicking on a product that appears in the movie leads, for example, to the product page on the eBay sales site.

Another possible application is augmented reality - for example, a smart phone owner could take a picture of a site in his environment and transmit it to the search system that finds all the relevant information about the site.

And another original idea that Dr. Bronstein and his colleagues are working on is to combine two biomedical images, for example a CT image and an MRI image into one image, even though the original images were made with different resolutions and with completely different means. Today, the researchers have managed to beat competing methods and they see great potential in the proposed algorithm for applications in medical imaging.

More of the topic in Hayadan: (Beresheet is the Hebrew name for the book of Genesis)

8 תגובות

  1. Thanks for the information, I hope the research will have uses beyond "warfare" in paratia.
    Today, when everyone has access to an excellent editing system, the mix/meshup can be made accessible and even legal.
    Like an open repository with the same "fingerprint" that will allow us to disassemble and assemble mixes of sound and image
    And that with the help of a mechanism of micro-payments you will allow the owners of the rights to receive their share.

  2. Tremendous development!! , I wonder if they will also be able to develop an algorithm that knows how to enlarge a film image or video to the size of a full screen when it is originally a tenth of that?

    In my opinion, it is possible if there is an algorithm that will identify the elements of the frame and magnify each part separately, and by connecting all the elements of the image, you will get an identical and high-quality image of the requested size, in which the eye will not be able to distinguish any difference!

  3. I once searched for a computer company by its logo that I found stuck on computer equipment, but the name did not appear and I thought to myself how nice it would be if Google, for example, had a service that allows you to do this type of search. So here this could be the solution. Instead of text, you upload an image in the Google search box (or in the above case a clip from a movie) and it finds the references.

  4. And regarding the idea of ​​combining a CT and MRI image, here is a similar idea that occurred to me years ago when I was still an armorer and spent whole nights observing. At first the device that worked on starlight amplification and then the thermal devices always had a certain lack of details that if there was a combination of the two forms of vision they would complement each other. So here's another idea to do. (Surely someone in the army has already implemented it)

Leave a Reply

Email will not be published. Required fields are marked *

This site uses Akismet to filter spam comments. More details about how the information from your response will be processed.