Open Source Video is a software project that attempts to translate motion picture files into Creative Commons licensed movies. Open Source Video treats a standard QuickTime movie file as a series of moving images. Then, by using the open source image recognition software known as pHash, Open Source Video tries to compare each frame of an "original" video to a database of Creative Common's photos downloaded off of Flickr. pHash then finds the best "match" between each "original" frame of a video and a CC licensed photo using a perceptual hash algorithm (akin to the types of algorithms sites such as Youtube.com use to detect copyright infringement).
Open Source Video is not meant to "replace" an original video. The semantic content of each frame (i.e., the objects you see in each frame) are ignored by a perceptual hash, thus what you see in a "original" video will often not "equate" to what you see in the Open Source Video. E.g., you see a cat in the original video, but you only see a dog in the Open Source version. By utilizing Flickr's tagged photo search, some layer of linguistic significance can be added to the database of Open Source images; however, this only feints at establishing a linguistic meaning between between two matched images.
This software treats a video as a series of moving images, one distinguishable from the next. Interesting enough, this is how Thomas Edison copyrighted the first motion picture in the United States ("Fred Ott's Sneeze", 1894). Edison printed each of the 45 frames of the "movie" and submitted this to the Library of Congress. By treating each frame as an individual image, we remove a video from a certain aesthetic layer and let it fall into a new reproductive medium, one which sees single pictures as interchangeable. From this point of analysis,the only difference between two images would be their issued license.
Open Source Video utilizes a copyright detection algorithm that is often used to detect unauthorized versions of a video from circulating online. However, when the pHash software is used in this context, it is only capable of detecting false frauds. By simplifying each image to a series of numbers that one can compare to any other image, we can quantify how fraudulent any picture or photo is in comparison to any other picture or photo, regardless of the image's context or where it fits in a larger discourse. By turning a copy detecting algorithm loose, it is only a matter of how large a comparison database can grow in order to increase the number and quality of frauds.
Thus, the ultimate goal of this project can be reached when a Vimeo.com algorithm, Vimeo.com being the domain where these Open Source video are uploaded, tags these Open Source videos as violating copyright.
For Sneeze: Searched for "Thomas Edison" and "Sneeze" on Flickr. Downloaded 422 images, paired them down to 83 images for my pHash image databank.
For Mickey Mouse: Searched for "Mickey Mouse" on Flickr. Downloaded 1239 images, paired them down to 450 images for my pHash image databank.
For Donald Duck: Searched for "Donald Duck" on Flickr. Downloaded 224 images, paired them down to 156 images for my pHash image databank.
For The Hangover: Searched for ANY CC licensed photo on Flickr. Downloaded 13211 images, used all images for my pHash image databank.
You can see the documentation to use the software at the Open Source Video Github Wiki (Under construction!)
You can download the source code from the Open Source Video Github (Under construction!)