Áëáë Øø Blockin Blockinòò Blockin

Color Invarian e Jan-Mark Geusebroek, Rein van den Boomgaard and Arnold Smeulders Intelligent Sensory Information Systems Department of Computer S ien e University of Amsterdam The Netherlands This paper presents the measurement of olored obje t re e tan e, under di erent, general assumptions regarding the imaging onditions. We exploit the Gaussian s ale-spa e paradigm for olor images to de ne a framework for the robust measurement of obje t re e tan e from olor images. Obje t re e tan e is derived from a physi al re e tan e model based on the Kubelka-Munk theory for olorant layers. Illumination and geometri al invariant properties are derived from the re e tan e model. Invarian e and dis riminative power of the olor invariants is experimentally investigated, showing the invariants to be su essful in dis ounting shadow, illumination, highlights, and noise. Extensive experiments show the di erent invariants to be highly dis riminative, while maintaining invarian e properties. The presented framework for olor measurement is well-founded in the physi s of olor as well as in measurement s ien e. Hen e the proposed invariants are onsidered more adequate for the measurement of invariant olor features than existing methods.