Near Infrared Face Image Recognition And Quality Assessment

Face recognition has been widely applied in access control, human-computerinterface, surveillance, public security and so on, due to its advantages of noncontact,high accuracy, stability and low-cost. However, the performance of the existing facerecognition is not robust to external disturbance, such as pose, expression andillumination variation. These problems need to be solved by the technique of facerecognition. In addition, along with the camera popularization, face recognitiontechnology is developed from the single background, static face image recognition tovideo sequences dynamic face recognition. As the video sequences are collectedrandomly, the quality of these face images varies considerably. To improve facerecognition rate, there is an urgent need for a mechanism can evaluate the quality ofthe input video sequence to choose the images which have higher image quality foridentification recognition. Considering that camera can capture20-30images persecond, to realize real-time processing with necessary accuracy is an issue that thetechnique of face recognition must be faced.Compared with the traditional visible face recognition, near infrared facerecognition has overcome the adverse effects of illumination variation. Based on nearinfrared face image, this paper has studied the three aspects of face detection, qualityevaluation and face recognition to improve the robustness, real-time and accuracy ofrecognition.For the Adaboost face detection just can detect the frontal face and be easilyaffected by the environment, we present face detection based on imagesegmentation.To realize coarse location, using the face region segmentation model atfirst to search the candidate region from the entire image. Then, evaluate the tilt of thecandidate region and correct the sloping one. Finally, the face position of the cuttingresults is located precisely by utilizing the Adaboost algorithm.For quality assessment of near infrared face image sequence, this paper presentsquality assessment based on multiple features fusion. The evaluation score of fivefeatures of sharpness, brightness, resolution, head pose and expression is computedindependently and then the final quality score is obtained by combining the scores offive features with weights.Centering about the problem of facial feature extraction and feature dimension reduction, this paper presents combining the Dual-Tree Complex Wavelet Transform(DT-CWT) and the Discrete Wavelet Transform (DWT) for face recognition. Firstly,the complementary of DT-CWT and DWT are used to extract the face features, forDT-CWT has the limited redundancy, approximate shift invariance and gooddirectional selectivity property, but it does not contain the horizontal and verticaldirections, while DWT can provide the direction of0oand90oscale description.Secondly, Locally Linear Embedding (SLLE) used to reduce dimension to realize thefast feature dimension reduction. SLLE is one nonlinear and non-iterativedimensionality reduction method which can preserve the local geometry ofhigh-dimensional data in the embedded space.Finally, this paper realizes the near infrared face detection, face recognition andquality assessment using OpenCV image processing library in the environment ofVisual C++, which basically meet the expected purpose and has very high practicalvalue.