Pixel Club: Improving Large-Scale Image Retrieval using Geometric Weighting

Speaker:
Dima Sezganov (EE, Technion)
Date:
Tuesday, 5.2.2013, 11:30
Place:
EE Meyer Building 1061

The Bag-Of-Features (BOF) approaches are becoming central in large-scale image retrieval. The standard BOF method is orderless, completely omitting geometric configuration of visual words. The geometrical information is usually involved only in the post-processing spatial verification step usually implemented with the RANdom SAmple Consensus (RANSAC) algorithm. To enable visual search in real-time, RANSAC can be applied only to a relatively small number of top candidates due to its computational requirements. In this work, we propose an alternative method to perform accurate spatial verification with a significantly lower computational cost. Experimental results show that the proposed method outperforms the baseline BOF, and achieves similar performance as RANSAC based spatial verification, despite the major difference in complexity.

Back to the index of events