Design And Implementation Of The Face Landmark System With Robust Expression

Human facial expressions as the carrier of human emotional expression is of great significance to the study of human mind. At present, facial expression recognition, micro expression detection technology have been widely used in clinical medicine, psychology, investigation and so on. In the process of facial expression recognition, the face landmark is the key to the final recognition result. In the recognition process, in addition to complex and changeful face expression, it has decorations, posture and light illumination changes. Therefore, robust face landmark technology in facial expression recognition process has important practical significance. In this paper, we focus on the research of robust face landmark algorithm, design and implement a demonstration system for positioning. The main research contents include the following three aspects:(1) In order to compare with the robust face landmark algorithm, the traditional Active Shape Model (ASM) algorithm is studied and implemented in this paper. This is a typical local constraint model,mainly includes two stages of training and search. Through the training of the model, the deformable model and local shape features for each landmark can be established, then searching around each landmark along the normal direction of the contour curve and judge the reasonableness of the shape, until convergence conditions are satisfied or optimal results is achieved.(2) This paper contrast and implementate three kinds of robust face landmark algorithms, respectively Robust Face Landmark Estimation Under Occlusion (RCPR), Cascaded Deformable Shape Model (CDSM) and Robust Discriminative Response Map Fitting (DRMF). RCPR algorithm is based on the CPR algorithm, and extracts feature through the shape index feature, then carries on the cascade regression training, at the same time to carry out the the face deformation and occlusion detection. To achieve robust, CDSM algorithm set the initial values of the area around the landmarks and multiple regional block combined as a group sparse model for feature extraction, then through the two cascaded deformation model with training.This algorithm has a significant effect in robust to changes in shape. DRMF algorithm through the parameter training response model with less parameters to represent the shape estimation, then update the convergence of the training samples to achieve more challenging training results, the algorithm has a significant effect in occlusion robustness.(3) With the AVEC expression library’s experimental analysis, summarize a conclusion:when the human face is deformed, the positioning effect of RCPR and CDSM are better, When the deformation degree is larger, the positioning effect of CDSM algorithm is better; when people with face occlusion, especially the eye occlusion, the positioning effect of RCPR and DRMF these two kinds of algorithm are better. In this paper, Robust face landmark fusion algorithm is designed, and a set of robust face landmark system is implemented which can compare two algorithms arbitrarily. With the positioning effect of the experimental data,the position of the landmarks can be derived from the system, so that it can be used to study the micro expression recognition algorithm.