3D Face Recognition Based On Clustering

Face recognition as a promising technology in the field of computer image, hasalways been a key problem to the outstanding research scholars. Although the level of the2D face recognition technology continues to mature, due to the impact of factors, such aslighting, expressions, gestures and other conditions, as well as the problem of the datainformation missing, so that the deviation seems unable to meet everyone’s needs in theprogress of society at the present. However, the3D data has the advantage of expressionof the facial features, and the3D data will be obtained more convenient and lower cost, sothat the domestic and foreign academic research focus shifted to the studies of3D facerecognition technology. So far, there has been emerged a large number of studies on3Dface recognition algorithm.In this paper, there are two feature vectors to be considered, one is the curvature thatis a good3D facial feature which can reflect the shape information of the surfaceproperties, the other one called ORB which has a strong competitive advantage. Thispaper will measure the similarity between facial models based on the curvature and theORB feature. Then, this paper will implement effectively a3D face recognition method toa large number of3D data set based on the agglomerative hierarchical clustering (AHC)algorithm.The main contents are as follows:1、A comprehensive review of the current research on3D face recognition will bedescribed, and it will be the guide to the following research.2、The features of ORB and curvature will be introduced in detail. ORB is a featureextraction algorithm which has comprehensive performance advantages especially inreal-time. This paper will use the Hamming distance to match the ORB feature points, andthen will get the final ORB feature vectors after removed the mismatch by RANSAC.For the depth image of the3D face, this paper will obtain the curvature feature by themethod of the local curve fitting based on the thinking of the least squares estimation.3、The3D facial features of the research objects contain extremely rich data,therefore, this paper will use a algorithm called agglomerative hierarchical clustering(AHC). Based on the similarity between facial models, the3D face recognition will beachieved successfully through the method of AHC by a region growing algorithm andsome validation indexes. Experiments show that the proposed method by AHC can obtain a higher recognition rate for the open database with certain robustness.