Research On Illumination Robust Face Recognition Technology

Face recognition technique is being gradually applied to real life and national security,many unsolved problems still exists,such as occlusion,illumination variation,expression and few samples.In this thesis,we focus on the key factors of illumination variation,and divide the illumination variation into two aspects: illumination intensity variation and illumination angle variation.For these two aspects of the targeted algorithm.The main focus of this thesis is research on illumination normalization and illumination invariant feature extraction algorithms and try to reduce the influence of illumination variation on face recognitionFirst we aim at the variation of illumination intensity,introduce and discusses several existed order features based on the illumination invariant feature extraction algorithm,Then experiments are implemented with illumination variants on AR and The Extend Yale B face database,results show that the algorithms based on order feature has strong illumination robustness.Then we aim at the variation of illumination angle,a normalization algorithm based on locally directional intensity mapping is proposed in this thesis.We have tested the performance of the above illumination normalization in the face database of Extended Yale B,the experimental results show that the recognition rate of the directional local intensity mapping under the large illumination angle variation was significantly higher than that of the traditional normalized algorithm.Finally,a method of illumination robust face recognition is proposed,which combines locally directional intensity mapping and ELTP features.An image normalization method based on locally directional intensity mapping is used to preprocess the image,and then extract ELTP feature,finally using the2? feature similarity measure and nearest neighbor criterion to realize human face image recognition.The experimental results show that the proposed algorithm for image illumination intensity and angle change at the same time compared to the existing algorithm has stronger robustness for illumination.