Statistical Model Based Facial Feature Point Location And Expression Recognition

Psychological research shows that about55%of the information is passed through facialexpressions in daily communications. Facial expression recognition has become one of theimportant research focuses in recent years. It has a wide range of applications inhuman-computer interaction and intelligent surveillance, etc. Along with the rapiddevelopment of computer science, facial expression recognition technology has madeconsiderable progress. However, there are still many unresolved problems and difficulties inbuilding a stable, accurate, robust and real-time facial expression recognition system.This thesis gives a detail description of the statistical modeling approaches: the ActiveShape Model. Active Shape Model is applied to locate facial feature points and the shapemodel parameters are used as the feature for recognizing facial expressions. This thesisproposes three improvements to the traditional Active Shape Model:1. The key points based initial shape estimation method is proposed. The center of left eye,right eye and mouth region of the face are selected as the key points to establish the bestaffine transformation between the average shape and initial shape estimation.2. Two-dimension local texture model of the normal and tangent direction for each featurepoint is established. In the model, the movable range of the feature point is expanded to aarea with a rectangular shape of the surrounding neighborhood in the search phase.3. In order to reduce the amount of computation and to accelerate the search speed, thesampling pixels on each side of the feature point is decreased along with increasing layersof the pyramid. This approach is named as the multi-resolution framework image search.Experimental results show that combining the key points based initial shape estimationmethod and the two-dimension local texture model can greatly reduce the number ofiterations, the searching time, and improve the accuracy of the feature point location and thefacial expression recognition.