3D Face Recognitionacross Pose Extremities

Pose-Invariant Face Recognition via RGB-D Images

“…We test our method on four publicly available datasets, including (1) high-quality RGB-D data from Bosphorus database and (2) some more challenging datasets that contain low-quality depth capture and illumination changing along with various expressions (e.g., CurtinFaces) and occlusions (e.g., Eurecom and Kiwi). We compare our method with the prior-arts methods including PGM [ 16 ], PGDP [ 17 , 26 ], DCT+SRC [ 11 ], COV+LBP [ 13 ], and RCRC [ 14 ]. We also evaluate in which way the 3D information can be fully explored and works better for 3D face recognition.…”

“…The algorithm has been compared with other state-of-the-art methods testing on this database. Reference [ 26 ] transformed the query 3D face from many different poses to frontal pose by using the Hausdroff Distance metric and then extracted the corresponding normal values for face recognition. Reference [ 26 ] achieved the average rank-1 recognition rate that is 66%.…”

“…Reference [ 26 ] transformed the query 3D face from many different poses to frontal pose by using the Hausdroff Distance metric and then extracted the corresponding normal values for face recognition. Reference [ 26 ] achieved the average rank-1 recognition rate that is 66%. It should be noted that [ 26 ] is based on 3D surface information, while the proposed method achieves an average accuracy of 87.1% on depth maps, outperforming the rank-1 recognition rate of [ 26 ] by 21.1%.…”

“…Reference [ 26 ] achieved the average rank-1 recognition rate that is 66%. It should be noted that [ 26 ] is based on 3D surface information, while the proposed method achieves an average accuracy of 87.1% on depth maps, outperforming the rank-1 recognition rate of [ 26 ] by 21.1%.…”