Image Recognition
With no way to insure the integrity and ownership of their works, artists and designers today often worry about their digital images being stolen via the internet. With the invention of Image Recognition Software, images can be tracked using fingerprint IDs from the original image.
Problem:
There are too many copies of images on the Internet, making it difficult for the creator to track their original work or find illegal copies of their work. Even though there are current watermark software and image databases that can be used to identify images, these techniques are unreliable. With these current technologies, any distortion to the images makes it nearly impossible to track or identify exact matches, and watermarks can be removed.
Solution:
This Image Recognition software uses an algorithm to create a specific fingerprint using points within the image. These points are not embedded, allowing the image to be identified despite distortion. This fingerprint is stored in a list and used when searching for an image match. When searching for a match, a fingerprint is created for the image in question. The fingerprint is used in a cross-correlation of possible images in order to match it to the original. The strength of the likeliness that the copy is from the original image (the “confidence”), is calculated with each set of identifying points between the two images, providing three possible outcomes: 1. If there is no confidence, the image is dismissed for lacking enough indentifying points to prove its affiliation with the original. 2. If the confidence is low, the system will output what matching similarities the images share, but is still not enough to prove that the images are the same. 3. If the confidence is high, the resolution of the image is changed and then cross-examined again. The end result produces an exact match of a copy to its original.
Advantages:
The advantage of this fingerprinting process, in comparison to current image recognition programs, is that it produces an exact match. With image databases, a great deal of overlap exists within the classification of images and image queries. The process of identifying copies is time consuming and inefficient due to this overlap and only produces similar but not exact matches of the originals. Another technique is watermarking; embedding a code containing owner, license, and tracking information within the image. Instead of searching for matching images, it searches only for the code. While watermarking may seem efficient, if the image is distorted the code may be lost. The process of Image Recognition is fast, reliable, and survives image distortions, making it ideal for tracking images over networks, computers, and/or databases.
If you’re interested in commercial opportunities for this technology, please contact the Office of Technology Transfer at: techtransfer.gmu.edu.
Interested in the latest patents at Mason? Click on the icon above, and subscribe to the Mason Technologies RSS Feed.