Improved supervised color constancy for color inspection

Abstract

In industrial applications, color inspection is difficult to be implemented because the environment illumination may vary unexpectedly, thus color constancy algorithm must be applied. Novak proposed the supervised color constancy thoughts, in which a few color chips of known spectral reflectance are placed in the scene to correct illumination changes. But his algorithm requires camera sensitivity functions which are usually difficult to obtain, and only numerical simulations is presented. In this paper, we proposed an improved supervised color constancy algorithm and applied it to color inspection. The algorithm need not to know the reflectance functions of color chips and the camera sensitivity functions, thus is more suitable for industrial uses. To ensure that this algorithm work well in industrial applications, imaging process of real system is studied and incremental-linear imaging model is adopted. Combined with this model, we gave the algorithm implementation on real systems, which shows satisfactory performance. Main thoughts and implementation details of the algorithm are presented in this paper, with experiment results and analysis.