Thermal Infrared Face Recognition Research Based On Physiological Structure

As for the face recognition under thermal infrared image, it not only have all the merits which near infrared face recognition have, but also have another important advantage, which, through the thermal image we could extract the facial physiological information. In this paper, we proposed a new face recognition method based upon the facial physiological pattern, and it is completely different from the traditionally geometrical based face recognition methods.The fundamental purpose of this research is to provide basic theory and methodology to the image segmentation, feature extraction and image matching based on the infrared facial image. Then, test the feasibility of the proposed algorithm, and investigate the robustness of its recognition rate over time. During the research, we first investigate the infrared image property in details, then, focus on the facial vascular network image segmentation. Finally, study the similarity between the vascular network and fingerprint, and bring the effective recognition methods of fingerprint recognition to face recognition, in order to develop a new face recognition method.At the image segmentation phase, we first adopt a Bayesian Segmentation framework to make the facial skin image segmented from the background. Then, after an Anisotropic Diffusion of image enhancement, we finally fulfilled the vascular network segmentation with a White Top Hat Segmentation algorithm.When it comes to the feature extraction and matching period, in this paper, we considered several fingerprint recognition methods. Firstly, we bring the One-Pass Parallel Thinning algorithm to guarantee the feature extraction process. Secondly, we make use of a Security Matching algorithm to perform the facial image matching. With the end of the proposed algorithm, we constructed experiments on the Notre Dame Biometric Database to evaluate our method, with C++ programming language under the OpenCv platform to realize our algorithm. As the experimental results show, our method is feasible and also its recognition rate over time is robust relatively.