Research Of Face Detection And Facial Feature Point Precise Localization

Face detection and facial feature points localization are both important in human facerecognition and computer vision system. Face detection is a critical first step of facerecognition system, facial feature points localization is the key which forms a connecting linkbetween the preceding and face recognition system, the positioning accuracy is directly relatedto the reliability of subsequent applications. This paper makes a deep research for facedetection and facial feature points localization, proposed a based on the haar-like feature andimproved AdaBoost algorithm, also present a facial features localization algorithm based onimage gray statistic, in view of the low resolution, small target human face image combiningbilinear interpolation algorithm to achieve face feature points location. The main works was asfollows:First, this paper presents the improved AdaBoost algorithm to extract optimal Haar-likefeatures to detect a single face and many faces. Regarding the internal feature method quickextract face haar-like feature, using this feature construction face classifiers, and set the samesample weights and thresholds initially, using the weak classifiers to classify, the correctsample weights decreased, the error sample weight is increased and standed out, according tothe classification results targeted for positive samples and negative samples were updatedweights and thresholds respectively, to find the optimal threshold for the weak classifier istrained, each cycle will be obtained and the optimal weak classifier to cascade, the finaloptimization of the composition of the final strong classifier.Secondly, the paper presents a precise positioning algorithm based on the combination ofimage gray statistic and distribution of facial features, realize a single face features location.Using facial features distribution, roughly determine the position of the core area of face (eyes,nose, mouth), then the use of gray-scale transformation, median filtering, edge detection imagepreprocessing methods, reduce the difficulty of the facial feature extraction. Finally, based onthe statistics of the gray facial features localization algorithm to achieve single face facialfeatures and contour point positioning.Finally, achieve a lower resolution, small target facial feature point positioning. Selectbilinear interpolation algorithm to enlarge the small target low-resolution face image tohigh-resolution images, combining a precise positioning algorithm based on the combination of image gray statistic and distribution of facial features, achieving a low-resolution smalltarget people face image feature point positioning, at the same time, verify whether to faceeach other based on facial features localization, effectively eliminate the phenomenon of falsedetection under complex and multi-face detection background.The experimental results show that the proposed algorithm single face and multiple facedetection rate can reach more than96%, applicable to complex facial expressions, with asignificant deflection angle, as well as different sizes of face detection.Meanwhile, the featurepoint positioning method used in this paper can be effectively applied to different sizes, faceimages with decorative objects (such as glasses, hats, collars, hair, etc.), with high positioningaccuracy, speed, and practical strong features.