Color estimation of illuminants has been applied to various applications such as the correction of white balance and color normalization for machine vision. The color estimation from a single image, however, is still challenging because each pixel value depends on three components: illumination, reflection, and characteristic of an image sensor, whose decomposition is a hard inverse problem. This research estimates the color of illuminant using color lines obtained from each super pixel, by assuming that the component of reflected light is locally constant. This article demonstrates the availability of this method for multiple illuminants of different colors, by experimentally showing a comparative evaluation against an existing state-of-the-art technology.
展开▼
