Abstract
Despite its perceptual significance, saturation is the least studied and least understood attribute of color. In CIELAB color space, it is calculated at the pixel level as a function of intensity (L*) and chroma (C*): C*/L*. However, within natural objects, chroma co-varies systematically with intensity in two different ways, depending on the nature of the interaction between light and surface. The body reflection is derived from the pigment composition of the object and varies mainly with shading; in the CIELAB L*-C* plane, the correlation between intensity and chroma is positive for this component: the brightest pixels are high in chroma and the darkest are low in chroma. The surface reflection results from the interface between the illuminant and the object’s outermost surface edges. It leads to a negative correlation between intensity and chroma: the brightest pixels have a low chroma, in line with the chromaticity of the illuminant. Here, we systematically varied the correlation between chroma and intensity to see whether this has effects on the perception of saturation. We used hyperspectral images of various fruit objects, as well as matte objects rendered using Mitsuba Renderer. For these objects, nearly all of the color variation was captured by the first principal component of all pixel values in CIELAB color space. We then manipulated the slope of this line, keeping basic chromatic statistics such as mean, variance, and range constant. Objects with a negative correlation between intensity and chroma were perceived to be much less saturated than those with a positive correlation. These results suggest that the perception of saturation cannot be explained by elementary statistics such as mean pixel saturation. Rather, the visual system estimates the underlying causes––illumination or pigmentation––and uses this implicit knowledge about the physics of light and surfaces to make judgments about invariant object properties.