"A meek endeavor to the triumph" by Sampath Jayarathna

Thursday, December 09, 2010

Reading #25: A Descriptor for Large Scale Image Retrieval Based on Sketched Feature Lines




              The main contribution of this work is a sketch-based query system for image databases containing millions of images. As most current retrieval algorithms for large image databases it is based on a small descriptor that captures essential properties of the images. A main feature is that it elegantly addresses the asymmetry between the binary user sketch on the one hand and the full color image on the other hand. The proposed descriptor is constructed in such a way that both the full color image and the sketch undergo exactly the same preprocessing steps to compute the descriptor. The resulting sketch based image retrieval system can be used by any novice user to quickly query the image database. The power of the system stems from exploiting the vast amount of existing images, which offsets obvious deficits in image descriptors and search.

The authors state that the descriptor’s performance is superior to a variant of the MPEG-7 edge histogram descriptor in a quantitative evaluation for which measured retrieval ranks of 27 sketches created from reference images out of the image database. 


            I’m totally perplexed. Is this related to sketch recognition? 


Francisco (Paco) Vides said...

I think it is. If you take into account that the descriptor is a feature of both the sketch and the images it is actually very similar to the various template matching recognizers seen before, just that in this case the submitted sketch is not really a sketch but an image and is not one but a million.

Jonathan H. said...

My answer is: yes. It's different in that it does not recognize the actual sketch like most of the other papers we've seen this semester. Instead, it's an image-based search system. I think it's about time someone created one of those.