Locally Adapted Detection and Correction of Unnatural Purple Colors in Images of Refractive Objects Taken by Digital Still Camera

We discovered significant error in color in images produced by a digital still camera used to capture scenes with a special setup of illumination and an object. The object is small, transparent, color-less and has light-refractive properties. Illumination is produced by a number of LEDs with known spectra. Due to light dispersion in the object, vivid monochromatic colored flares can be seen and captured. Images from a digital still camera captured in the described setup occasionally exhibit bright purple (almost pink) colors, which do not correspond to any monochromatic color. In this paper, we analyze the origins of this effect by examining different properties of the setup and analyzing RAW images. We propose a simple and efficient algorithm for correction of those unnatural purple colors by using only final JPEG image produced by the camera. In this algorithm, we develop a continuous transform which maps all unnatural colors to the natural ones in a perceptually uniform color space. Then, a 3D look-up table is constructed using this transform, which in turn is used for further correction of images captured by the camera.