Frontal Face Image Quality Assessment

With the rapid development of the multimedia and sensor technology, thebiometric authentication system has been born. Face recognition as an effectivebiometric authentication approach gains wide attention of the academia and industry,which becomes one of the most important research subjects in pattern recognition andcomputer vision. And face recognition has been widely used in real life, such as accesscontrol system, attendance management system and ID authentication. However,because of the nonstandard operation in acquisition and image compression, the frontalface image which is the most common authentication information carrier will bedistorted more or less. All these distortions affect the quality of the face image seriously,and increase the difficulty of face biometric authentication inevitably. Thus, it is of greatpractical significance to evaluate the quality of frontal face images.The status of eyes and mouth can be considered as the most important factorswhich affect human expression. Based on this fact, a method based on sparserepresentation is proposed for unnatural expression detection. The object detection andfeature points location are used to segment the eyes and mouth region from frontal faceimage, the status of eyes and mouth can be detected via sparse representation, then thehuman expression can be estimated qualitatively. Experimental results illustrate that theproposed method achieves good performances.The thesis proposed a quality assessment metric based on statistical analysis forlocal illumination in frontal face images. According to a large number of experimentaldata, a quality assessment model is proposed for global illumination. The biologicalcharacteristics of the white of the eye are utilized to evaluate the color quality in frontalface image. Experimental results demonstrate that the proposed methods have goodconsistency with subjective perception values and the objective assessment results canwell reflect the illumination quality of frontal face images.Moreover, in the VC++development environment, the frontal face image qualityassessment platform is successfully constructed with many of the classic qualityassessment methods integrated into it. And this platform bears great utility value, whichwill do favor to the popularization of the frontal face image quality assessment methods.