Curvature-based face surface recognition using spherical correlation-principal directions for curved object recognition

Abstract

Surface curvat.ures such as Gaussian: mean and principal curvat,urcs are intrinsic surface propert,ies and havc played import,ant, roles in curved surface analysis.in this pq'er, we present a correlat,ion-h;lsecI facc rccognition approach based on the analysis of maximum and minimum principal curvaturcs and t,hcir tlirect,ions. \Ye t,rea.t, face rccognition problem as a 311sha.pe recognit,ion problem of free-form curved surfaces.Our approach is based on a 311 vect,or sets correlat,ion mtt,hod which does not. require eit,her face feature cxtrwtion or siirface segmenhtion. Each face in bot,h input images and the inodel datalmse, is represcntetl a.s an Extendcd Gaussian Image( EGI), construct,ed by m a.p pin g p ri n ci p a1 cu rvat u res and t heir di re c t ion s at e x h surface point,s, onto t x o unit, spheres, each of rvhicti represent.s ridge a n d valley lines respectively. Individual face is t,tien recognized by evaluating t,lie simila.rit.ies among others by using Fisher's spherical correlation on EGI's of faces.The method is tzest,ed for it,s simplicit,y and robust,ness a.nd successively implcmentcd for cach of f x c range images from NK.CC( National Ktsearch Couiicil Chnada) 31) ima.ge d a h files. Results shows t,tia.t.shape informa,t.ion from surface curva.t,ures provides vital cues in dist,inguishing and identifying such fine surface structure a s human faces.

3D Face Recognition: Technology and Applications

“…They have been used for facial surface segmentation [66] as well as representation [90]. Commonly used descriptors are maximum and minimum principal directions [90], and normal maps [3,4].…”

Soft Computing Based Range Facial Recognition Using Eigenface

“…Many researchers have used 3D face recognition using differential geometry tools for the computation of curvature [9]. Hiromi et al [10] treated 3D shape recognition problem of rigid freeform surfaces. Each face in the input images and model database is represented as an Extended Gaussian Image (EGI), constructed by mapping principal curvatures and their directions.…”

Matching 2.5D Scans for Face Recognition

“…A few approaches utilized depth information provided by 2.5D range images [2][3][4][5], but most efforts have been devoted to face recognition from only two-dimensional (2D) images [1]. Current 2D face recognition systems can achieve good performance in constrained environments, however, they still encounter difficulties in handling large variations in pose and illumination [6].…”