Yissum, partnered with the Hebrew University of Jerusalem published (IEEE) information on a new algorithm for ultra-rapid image retrieval which they say is robust and reliable. It is obviously going to be very useful in large image databases. They say that basically you could take a photo of a restaurant and then immediately get information on it from the photo analysis. In fact it has many applications, they give the example of surveillance cameras.
The algorithm decides which pattern and areas (windows) are non-similar rather than establishing a distance measure. This way unnecessary information is not computed so it's faster. They say it can detect "low quality patterns, rotated patterns or patterns that are partly occluded."
From their website:
Through our support and encouragement of research, development, and education, we are dedicated to turning science into commercial products for society’s use and benefit.
Goals & Objectives
- To protect, promote and market commercially promising inventions and know-how developed at the Hebrew University of Jerusalem
- To find the “right fit” for each intellectual property asset in our portfolio
- To deliver, manage, and optimize knowledge transfer to the global market through a variety of business development activities and services
This will be very interesting to any image search engine in particular, because their databases are large and they often rely on tags. There is work in image recognition for sure, but this really cool because I'd quite like to pick a photo and immediately get information on it. This would be particularly powerful in mobile search.