Yeda Company signed an agreement with Adobe Company on a method for summarizing visual information developed at the Weizmann Institute of Science

The two-way similitude method, developed by Prof. Michal Irani and the research students from her group (then-now researchers in their own right) from the Department of Computer Science and Applied Mathematics at the Weizmann Institute of Science, is a technique for summarizing visual information, both images and videos

The "Knowledge Research and Development" company, the commercial arm of the Weizmann Institute of Science, announces today the signing of a license agreement with the international software company Adobe, on a method for summarizing visual information, based on a two-way similarity index, developed at the Weizmann Institute of Science.

The two-way similitude method, developed by Prof. Michal Irani and the research students from her group (at the time) Dr. Denis Simkov, Dr. Yaron Caspi and Dr. Eli Shechtman, from the Department of Computer Science and Applied Mathematics at the Weizmann Institute of Science, is a technique for summarizing visual information, images and videos as well. Unlike other methods, which "cut" the image or shrink it, or - in the case of a video - produce a short and partial clip, the new method produces a visual essence that is complete and coherent. It is actually a reduced or shortened version of the original, which retains the most relevant information. The two-way feature of the method ensures that the resulting image is visually coherent, that is, visually logical, and as pleasing to the eye as the original. In contrast to cutting or cropping, in which important information may be lost, or to reduction, in which the resolution is lost, in the two-way compression method, both the important information and the resolution details are preserved, despite the size change.

The method is based on deleting redundant and redundant information from the image or video. So, for example, a picture of a car driving on a city road, after compression, will still include the car in its entirety, but will only include part of the road and some of the houses. The algorithm developed by Prof. Irani will identify the rest of the road, as well as the segments that contain houses, as repeating elements. Extracting videos works in a similar way, except that the software detects duplicates in time-space information. A gradual process of reduction and control ensures that the final result will be coherent and without visible "seams".

In addition to extracting images and videos, the new method may have additional applications, including completing missing sections in images and videos; Creating montages (combinations) from several different images; rearranging information in pictures and movies (for example, changing places of objects in the picture); automatic cutting; Image synthesis (ie expansion of an image, instead of its extracts); and "morphing" (producing a video sequence by creating a smooth transition between two images, even if there is no connection between them).