Specularities on the facial images have a great impact on the performance of face area segmentation, detection and recognition. In this paper, a novel automatic specularity detection and removal method is proposed. First potential specularities are detected using PI diagram which is constructed by the principal vector of image's RGB vectors and illumination intensity. Second the full areas are got by the inner-growth and outer-growth algorithms. The last step, the specularities are eliminated by a template with weight. This method requires no interactive operation, be realize simply and is robust to noise.
展开▼