One of the liabilities of the internet, copyright
violation, could be solved by an invention from Japanese
boffins.
A Waseda University
research group has developed a fast way to search through large
numbers of images to find close matches to specific
photographs.
The system was developed to help identify copyrighted images
being used without permission on the internet. Another spin off is
that the technology could be used to search for images that have
been altered.
In one test using the new search system, a target photo could be
detected from a random collection of 480,000 images in roughly five
seconds.
The system uses a two step process to identify images that are
similar to an original, and then decide whether they have been
copied. In stage one, the colours and line angles of the target
photo are used as clues to search through collections of images to
find those that are generally similar in appearance. Stage two
finds the machine dividing the target photo into five partially
overlapping regions on which the search is repeated for each. For
every "hit" on each region (of the image) a corroborating set of
investigations is then initiated on other parts of the photo to see
if the entire image. If they all match, then copyright is likely to
have taken been violated.
