Independent Gabor analysis of illumination insensitive isophote-angle for face recognition

A new independent Gabor analysis of the isophote angle is proposed for face recognition. The isophote is the curve of constant intensity in image, which represents the geometric object contour. Isophote angles are proved to be insensitive to the illumination; therefore, they are adopted in our method for recognition under low-level lighting conditions. Then, a bank of Gabor filters is built to exact Gabor representations of the isophote angles. Last, the higher-order statistical relationships among variables are extracted for classifier, where the cosine distance function computed in gradient domain achieves the best performance. The experimental results demonstrate that our method outperform the conventional methods under varying illumination and other outliers (pose, expression, distance, etc.), especially in the large scale face databases.