Research On ASEF And Pictorial Structures Based Facial Landmark Detection

Facial landmark detection as an important prerequisite on face study, it can provide basis data for post-processing on face, and can be widely used in face recognition, facial expression recognition, fatigue driving and other fields. In recent years, with the improvement of society and technology, face recognition technology has been widely applied in social security and surveillance systems. The demand for effective automatic verification is increasingly urgent, which makes facial landmark detection attract increasingly attention. Until now, domestic and foreign researchers have proposed several methods of facial landmark detection, which will be divided into local landmark detection and global landmark detection.Common local landmark detection method include the Hough transform, integral projection curve and ASEF (Average of Synthetic Exact Filters), these methods always use the priori knowledge to locating single face organ. ASEF as a local landmark detection, it has exhibited its impressive performance in application of eyes detection. However, the accuracy of ASEF will deteriorate severely when the number of landmark is large. One reason is that the relative position of facial landmarks is not intrinsic considered in ASEF. Common global landmark detection method include AAM (Active Appearance Model), ASM (Active Shape Model) and pictorial structures model. Global method can locate a full face or multiple organs at one time. Pictorial structures as a model-based approach, the global knowledge of landmark distribution can be efficiently considered, but it works not very well at extract features of organs.In this paper, we propose a method which can comprehensive the advantages of local and global detection method. Paper’s main work is as follows:1) System elaborated the history and status of research on facial landmark de-tection. Detail kinks of landmark detection approaches, including gray-level-based algorithm, Geometry-based algorithm, knowledge-based algorithm, stat-istic-based algorithm and wavelet-base algorithm.2) Research on algorithms of locating rough position of special facial organs. Introduce eyes locating methods such as Hough Transform algorithm, edge ch-aracter analysis algorithm and ASEF algorithm. Nose location methods such as Subclass Discriminant Analysis algorithm. And two kinds of mouth localiza- tion methods.3) Introduce two model-based approaches on facial landmark localization:Pictor-ial Structures method and ASM method.4) Proposed the improved algorithms based on ASEF and Pictorial Structures. First, by consider of the affect of a on facial landmark localization, we propose an improvement by adding Simulated Annealing algorithm to ASEF. Anoth-er improvement scheme is the combination of ASEF and Pictorial Structures. A-nd the experiments prove the feasibility of the two algorithms proposed.