The usage of color invariance in SURF

Gang Meng,Zhiguo Jiang,Danpei Zhao
DOI: https://doi.org/10.1117/12.833509
2009-01-01
Abstract:SURF (Scale Invariant Feature Transform) is a robust local invariant feature descriptor. However, SURF is mainly designed for gray images. In order to make use of the information provided by color (mainly RGB channels), this paper presents a novel colored local invariant feature descriptor, CISURF (Color Invariance based SURF). The proposed approach builds the descriptors in a color invariant space, which stems from Kubelka-Munk model and provides more valuable information than the gray space. Compared with the conventional SURF and SIFT descriptors, the experimental results show that descriptors created by CISURF is more robust to the circumstance changes such as the illumination direction, illumination intensity, and the viewpoints, and are more suitable for the deep space background objects. © 2009 Copyright SPIE - The International Society for Optical Engineering.
What problem does this paper attempt to address?