Facial Geometry Identification through Fuzzy Patterns with RGBD Sensor

Abstract

Automatic human facial recognition is an important and complicated task; it is necessary to design algorithms capable of recognizing the constant patterns in the face and to use computing resources efficiently. In this paper we present a novel algorithm to recognize the human face in real time; the system's input is the depth and color data from the Microsoft KinectTM device. The algorithm recognizes patterns/shapes on the point cloud topography. The template of the face is based in facial geometry; the forensic theory classifies the human face with respect to constant patterns: cephalometric points, lines, and areas of the face. The topography, relative position, and symmetry are directly related to the craniometric points. The similarity between a point cloud cluster and a pattern description is measured by a fuzzy pattern theory algorithm. The face identification is composed by two phases: the first phase calculates the face pattern hypothesis of the facial points, configures each point shape, the related location in the areas, and lines of the face. Then, in the second phase, the algorithm performs a search on these face point configurations.

RGB‐D face recognition using LBP with suitable feature dimension of depth image

“…Since the depth map returned by RGB-D sensors is not as precise as a 3D sensor, the existing 3D face recognition approaches may not be directly applied to RGB-D images. In contrast, RGB-D images have been used for several computer vision tasks such as object tracking, face detection, gender recognition, face recognition, and visual imitation [2,[17][18][19][20][21][22][23]. To the best of our knowledge, there is no reported research on the recognition of faces using the LBP texture features of RGB and depth images and the error correcting output code (ECOC)-based multiclass support vector machines (MSVMs).…”