H. Fillbrandt and K.-F. Kraiss (Germany)
Illumination Normalization, Color Constancy, Color-based Object Recognition.
One of the most demanding problems in computer vision today is robustness under changing illumination. In con trast to the common bottom-up approaches of illumination normalization by color constancy algorithms, we propose a top-down approach to normalize the image data by match ing it to the color distribution of the respective reference object. The method uses a flexible cluster-based hierarchi cal model that is based on color theoretical facts as well as on the observation that the qualitative topology of the data distribution in color space is more important than the ac tual color values. We show the successful application of this approach to the task of illumination invariant object recognition.
Important Links:
Go Back
