Illuminant Estimation from Projections on the Planckian Locus

Abstract

This paper deals with the automatic evaluation of the illuminant from a color photography. While many methods have been developed over the last years, this problem is still open since no method builds on hypotheses that are universal enough to deal with all possible situations. The proposed approach relies on a physical assumption about the possible set of illuminants and on the selection of grey pixels. Namely, a subset of pixels is automatically selected, which is then projected on the Planckian locus. Then, a simple voting procedure yields a robust estimation of the illuminant. As shown by experiments on two classical databases, the method offers state of the art performances among learning-free methods, at a reasonable computational cost.

Estimating illuminant chromaticity with a low-power color pixel

“…In this work, we consider the chromaticity estimation algorithm recently proposed in [9], because (1) it works on the 978-1-4799-8591-3/15/$31.00 ©2015 IEEE chromaticity coordinates uv, that are invariant to changes of light intensity; (2) it does not require any a priori information about the device and/or the illuminant; (3) it is able to estimate multiple illuminants possibly present in the viewed scene; (4) it is suitable to be implemented on a sensor. Fig.…”

“…In [9], the distribution D is weighted by the pixel brightness to limit the influence of the dark pixels, whose contribution usually adversely affects the chromaticity estimation. We do not implement this weighting procedure in our sensor: in fact, thanks to the auto-exposure capability, the sensor acquires the color signal in the best luminance condition.…”

“…In case of multiple illuminants, the number of the illuminants and their CCTs are estimated by computing the modes of the distribution D. For more details we refer to [9].…”

“…We remind that the CCT of a pixel with chromaticity (u, v) is the temperature of a blackbody radiator having chromaticity (u, v) and the CCT of a gray pixel is the CCT of the light illuminating that pixel. In [9], the CCT T of a gray pixel is computed from its uv coordinates by Robert's method [8]. Usually, only the values of CCT between 2000K and 20000K are retained, because lower and higher CCTs are highly improbable in natural scenes.…”

“…Precisely, (i) this sensor captures a set of color signals from different locations in the scene through a low-power color pixel recently developed [7], (ii) a microcontroller interfaced with the pixel selects the signals corresponding to gray pixels and computes their correlated color temperature (CCT) by Robert's method [8]; (iii) the CCT's computed before are sent to a processor, that estimates the light chromaticity from the most frequent CCT according to the algorithm described in [9].…”