Spectro-spatial gradients for color-based object recognition and indexing

This paper presents illumination pose- and illumination color-invariant color feature descriptors for object recognition and indexing which are derived from spectral (color) and spatial derivatives of logarithmic image irradiance. While the use of spatial gradients and spatial ratios of image irradiance have been suggested for limited viewing-pose invariance and illumination-color invariance, respectively, gradients in the spectral direction and combination of spectral and spatial gradients have not been fully investigated. We present a unified framework for analyzing spatial and spectral gradients of logarithmic image irradiance, and suggest that spectro-spatial gradients have rich potential for developing local and global descriptors of object color. Experimental results are presented to demonstrate the efficacy of the proposed descriptors.