Image categorization using color G-SURF invariant to light intensity

Citations of this article
Mendeley users who have this article in their library.


The task of unsupervised image categorization is not easy. Issues include a wide range of rotations, shifts, scales, and variability of light intensity in images. Methods, which provide invariance to geometric and light intensity distortions, are in the area of interest. Our contribution deals with a family of novel descriptors joining geometric, color, and light intensity specialties in a neighborhood of feature point called as color Gauge Speeded Up Robust Feature (G-SURF). The color G-SURF extraction does not applied to the whole annotated image, but concern the preliminary segmented image using J-SEG algorithm. Only 5-7 large area segments are involved in the categorization procedure. This subset is enough for good image categorization based on Support Vector Machine (SVM). A set of eight scene categories including 2,688 images with sizes 256 × 256 pixels from dataset represented by Oliva and Torralba was used for experiments. The proposed descriptors provide good results in unsupervised image categorization achieving the precision values up 80-98%.




Favorskaya, M., & Proskurin, A. (2015). Image categorization using color G-SURF invariant to light intensity. In Procedia Computer Science (Vol. 60, pp. 681–690). Elsevier B.V.

Register to see more suggestions

Mendeley helps you to discover research relevant for your work.

Already have an account?

Save time finding and organizing research with Mendeley

Sign up for free