Research Of Facial Feature Landmark Localization Based On ASM And POEM

Face recognition, as one of the significant biometric feature technology, has recently been widely studied in recent years, due to its quick and convenient process, visual effects, and simple equipment. The facial feature landmark localization is the basis of face recognition method, which directly affects the accuracy of the positioning result of face recognition. With further research, facial feature landmark localization methods are also applied to the face reconstruction, expression recognition, analysis of human psychology as well as driver fatigue analysis, and therefore have an important significance. Active shape model, as one of the facial feature landmark localization methods, has been widely used because of its quick speed and localization accuracy. However, the effects of localization are easily influenced by the initialized shape and some other external influences.Based on the ASM, aiming to solve the problem of localization while facing with the pose, light and expression variations, some improvements have been proposed:First, the frontal, left-side and right-side model are established to discriminatively represent faces with different poses in the training phase. And then Model Selection Factor is proposed to automatically select the most suitable model in the process of search, achieving a reliable initialization for the face.Then, POEM is used to replace local texture model of ASM so that we can achieve the best position of each landmark and avoid the influences of complicated nonlinear change and large deformation to much extent.Thirdly, the worst localization organ is achieved by computing mean error of each face component, therefore, a second localization is presented to discriminatively refine the subtle shape variation for some organs and contours.Experiments are conducted in four main face datasets(IMM?CMU PIE?BioID and outdoor face dataset LFW), and the results demonstrate that the proposed method accurately localizes facial landmarks and outperforms other state-of-the-art methods, especially in the condition of pose, light and expression variations.