Automatic Color Profiling of Digital Cameras Using Unordered Photos

We present a novel approach for automatic color profiling of multiple cameras into a single RGB color space using only unordered photos of a scene as input. By performing bundle adjustment, camera poses and image features are automatically detected and used to sample RGB values of corresponding image features. Retrieved RGB values are used to model the relationship between the RGB color spaces of different cameras using linear least square fitting. After color profiling all color output is transformed into the color space of a reference camera. This greatly simplifies characterization as manual characterization steps are only needed for the reference camera.