Colored Illumination for Enhancing Discriminability in Machine Vision

Abstract We describe how to design a colored illumination that maximizes the discriminability of objects in red-green-blue images and we describe the underlying mathematical theory. This procedure is useful in applications where the illumination can be controlled, e.g., automated inspection. Our technique finds the optimal colored illumination in terms of the camera's spectral response and the objects' spectral reflectance. An experiment with filters simulating colored illumination on painted color patches is reported. This experiment confirms the effectiveness of our design technique. Approximations for the spectral reflectances and the illumination sources are useful for designing real colored illumination. These approximations were tested successfully on live potato plantlets from tissue culture. These experiments on color patches and potato plantlets demonstrate the usefulness of properly designed colored illumination in machine vision.