An interpolation based approach for lighting variation in image recognition

We present an interpolation based approach to model the variation of illumination in images. Our approach allows for both synthesis and analysis of images in different lighting conditions. Hence, our model is a compact representation of the set of all images with the lighting conditions defined within a parametric space. Based on this approach, we propose image recognition methods to consider varying illumination conditions. We evaluate our recognition methods with an object image dataset and with the Yale extended face database. With our object image dataset we obtained a 99.53% recognition rate. The face recognition method proved to be robust and performed on par with the current state of the art algorithms in terms of recognition rate. We recorded an overall 91.92% recognition rate by testing every single image (including the ones with extreme lighting conditions) against the rest of the dataset. Furthermore, our model is general and can be extended to handle geometry and any other imaging parameters.