Abstract: To estimate the illumination of outdoor scenes captured under a fixed viewpoint,we first extract image regions which are sensitive to the scene illumination at offline stage.Then we take the mean value and the unbiased estimation variance of sensitive regions as statistical parameters and construct the model between light parameters and statistical parameters using partial least squares regression analysis.At online stage,we use the constructed model to estimate the illumination of every frame in real time.Experiments demonstrate the effectiveness of our algorithm.