Pose-Invariant Face Recognition via RGB-D Images

Figure 1

Framework of the proposed method. The input depth map is first frontalized and registered to a predefined reference model. Its RGB counterpart is then aligned to the database facing the same direction as the query. Finally, different classifiers are used for both texture and depth, and the final result is fused at score level. Here, ICP denotes the iterative closest points algorithm, and and are the rotation and translation matrixes estimated by ICP for registering the query face.