Metamer mismatching and its implications

Due to the phenomenon of metameric mismatching, Logvinenko [PLoS One, 2015 Sept. 10, 10(9)] has pointed out that color is not an intrinsic property of an object. Not only can a change in the illumination lead to a substantial change in the measured or perceived color of an object, two objects that are metamers (i.e., their color signals match under one illuminant) may not match under a different illuminant. This phenomenon is referred to as metamer mismatching. Given a surface reflectance illuminated by a given light, there can be many other surface reflectances for which the eye provides an identical LMS cone-triple response or an identical RGB response in the case of a camera. The tristimulus color values of these metamer reflectances can disperse into many different tristimulus values under a different illuminant. The set of all such possible tristimulus values defines a convex hull known as a Metamer Mismatch Body (MMB). Metamer mismatching poses several challenges in color-based machine vision, such as color prediction, color discrimination variability and the color accuracy of a digital camera. In this work, a variety of approaches to measuring the extent of metamer mismatching and partially mitigating its effects are considered. We begin by investigating the performance of existing color prediction methods in predicting the results of asymmetric color matching experiments (i.e., finding the least-dissimilar matching pairs of colored papers under different illuminants). Because of the possibility of metamer mismatching, it is a mistake to interpret one answer as the 'correct' answer. However, we demonstrate that all the computational methods studied are not capturing some important aspects of the observers' least-dissimilar matching strategy. MMBs are 3D volumes that tend to be quite wing-like in shape. By modeling the MMB by its equivalent (in terms of its inertial moments) ellipsoid, a new metric for evaluating the colorimetric accuracy of digital color cameras is proposed. The advantage of the new metric is that it is based on a theoretical principle rather than simply computing the average error over a chosen set of representative test reflectances. Furthermore, a theory about the inverse relationship between color discrimination ellipsoids and the extent of metamer mismatching is proposed. Statistical analysis over the existing datasets provides evidence that metamer mismatching can possibly explain why color discrimination varies throughout color space as it does. It is shown that the proposed theory paves the way to predict the Just Noticeable Differences (JNDs) in different regions of a color space.