The Research On Conditional Regression Forests For Facial Feature Points Detection And Application

Accurate facial feature points location is one of the important processes in the application of analysis of facial data and face recognition.Face annotation aims to accurately estimate the spatial locations for face images under varying poses,expressions and illuminations.Meanwhile,from the complex environment precise positioning of facial feature points is a very difficult task.Hence,accurate face annotation is a challenging research problem.Our research has analysed the problems in detail which will be met in application.The dissertation studies an improved conditional regression forests for facial feature points detection.We firstly introduce the target tracking algorithm.We must track the face in the facial images,detecting facial feature points location.In this work,we present a method based on regression forests that detects facial feature points in real-time.Since regression forests learn the spatial relations between image patches and facial features from the complete training set and average the spatial distributions over all trees in the forests tend to introduce a bias to the mean face.This is very problematic for facial feature detection since subtle deformations affect only the image appearance in the neighborhood of a specific feature point.In order to steer the impact of patches close to a facial feature,which adapt better to local deformations but are more sensitive to occlusions,and more distant patches that favor the mean face,we introduce an objective function that allows to find a good trade-off.Finally,we can get robust result of facial feature detection.Another contribution of this work is the introduction of conditional regression forests.In general,regression forests aim to learn the probability over the parameter space given a face image from the entire training set,where each tree is trained on a randomly sub-sampled training set to avoid over-fitting.Conditional regression forests aim to learn several conditional probabilities over the parameter space instead.The motivation is that conditional probabilities are easier to learn since the trees do not have to deal with all facial variations in the appearance and shape.Since some variations depend on global properties of the face like the head pose,we can learn the trees conditional to the global face properties,the probability of the head pose is estimated from the image and the corresponding trees are selected to predict the facial features.In this way,the trees that are selected for detecting facial features might vary from image to image.Finally,we have developed the system of locating facial feature points using conditional regression forests.Through designing each function module rationally and adopting Microsoft Visual 2010 tool to carry on system development.The system developed is suitable for developing the application software in future.