3D Face Recognition Based On LBP

Human being’s curiosity to explore the unknown things is the driving force to advance science and technology. Nowadays, having been integrated into all walks of life, science and technology changes people’s living habits gradually. Among them,face recognition as the representative of biometric technology plays a very important role in many fields, such as: criminal investigation, online payment, access control systems and so on. And in these areas, it has important application merits. Although the 2D face recognition technology entered its mature stage, due to the limitation of the volume of 2D data and the interference form the light, posture, facial expressions and other external factors at the time of 2D data acquisition, its development was hindered. By contrast, the 3D face data not only includes a wealth of information in3 D space, but also is less affected by external factors while collecting the data.Therefore, the emphasis of 3D face recognition research gradually shifted into 3D field.Local Binary Pattern(LBP) is an excellent method of parts-based feature extraction, which has the advantages of extraction accuracy and insensitive to light.In this paper, several advantages of the current face recognition method was absorbed,and on this basis, a fusion based on multi-dimensional, three-dimensional human face multi-scale LBP feature vector information using ridge regression algorithm to train the classifier was proposed, and instead of using feature vector, we adopt the mark vector to classify in 3D face recognition. In this paper, the main work are as follows:1、Pretreatment. Using Ada Boost classifier for face detection and extract the face region in the form of 2D images; Cutting the 3D face model, and then obtaining the3 D facial depth image by projecting its face area.2、Feature extraction. 3D facial feature vector consists two parts, one part concludes the 3D facial depth image LBP feature vector; the other part contains the texture image corresponding to the 2D face of multi-scale LBP feature vector. Firstly,we obtain the 2D face image by using different scales of wavelet decomposition.Secondly, through the way of LBP operator in Uniform mode, we extract the feature vector of depth image and the multi-scale image. And finally, cascade the fused multi-dimensional, multi-scale 3D face information LBP Feature vector.3、Classification and identification. To reduce the computational complexity,accelerate the recognition speed, a method of labeling vector estimation based on4、feature vectors for face recognition was adopted. The utilization of PCA can achieve dimensionality reduction of 3D facial feature vector. And then reuse ridge regression algorithm training feature vector matrix and vector matrix to obtain labeled ridge regression projection matrix, and finally face recognition feature vectors to the projection matrix obtained marked vector estimation value, calculated the estimated value between vectors and labels face recognition.Experiments show that this method can effectively extract the 3D facial features and it has a high recognition rate, has a positive effect to the development of 3D face recognition.