Global context descriptors are vectors of additional information appended to an existing descriptor, and are computed as a log-polar histogram of nearby curvature values. These have been proposed in the past to make Scale Invariant Feature Transform (SIFT) matching more robust. This additional information improved matching results especially for images with repetitive features. We propose a similar global context descriptor for Speeded Up Robust Features (SURFs) and Maximally Stable Extremal Regions (MSERs). Our experiments show some improvement for SURFs when using the global context, and much improvement for MSER.

doi.org/10.1109/CRV.2010.47
7th Canadian Conference on Computer and Robot Vision, CRV 2010
School of Computer Science

Carmichael, G. (Gail), Laganière, R. (Robert), & Bose, P. (2010). Global context descriptors for SURF and MSER feature descriptors. Presented at the 7th Canadian Conference on Computer and Robot Vision, CRV 2010. doi:10.1109/CRV.2010.47