With the development of consumer-level depth sensors, 3D face point cloud data can be easily captured now. However, such data are often accompanied by low resolution, noise, and holes. At the same time, high-precision 3D scanners are bulky and can not be widely used in daily applications due to costs and inconvenience. To fill the gap between low and high resolution 3D faces, we propose a two-stage framework named the face point cloud super-resolution network (FPSRN) to recover high-resolution 3D face data from the low-resolution counterparts. As the human faces can be aligned into a unified coordinate system, we formulate point cloud super-resolution as a z-coordinate prediction problem. Cascaded auto-encoders are employed to retain both global structure and boundary information of different face regions during super-resolution. Compared with state- of-the-art point cloud completion methods and depth estimation methods, our method improves the Earth-Mover’s Distance (EMD) and the Root Mean Square Error (RMSE) metrics by 43% and 25%, respectively.

This paper proposes an attention-guided feature disentangling framework (AgFD) to eliminate the large cross-modality discrepancy for Heterogeneous Face Recognition (HFR). Existing HFR methods either focus only on extracting identity features or impose linear/no independence constraints on the decomposed components. Instead, our AgFD disentangles the facial representation and forces intrinsic independence between identity features and identity-irrelevant variations. To this end, an Attention-based Residual Decomposition Module (AbRDM) and an Adversarial Decorrelation Module (ADM) are presented. AbRDM provides hierarchical complementary feature disentanglement, while ADM is introduced for decorrelation learning. Extensive experiments on the challenging CASIA NIR-VIS 2.0 Database, Oulu-CASIA NIR&VIS Database, BUAA-VisNir Database, and IIIT-D Viewed Sketch Database demonstrate the generalization ability and competitive performance of the proposed method.

Dictionary based orientation field estimation approach has shown promising performance for latent fingerprints. In this paper, we seek to exploit stronger prior knowledge of fingerprints in order to further improve the performance. Realizing that ridge orientations at different locations of fingerprints have different characteristics, we propose a localized dictionaries-based orientation field estimation algorithm, in which noisy orientation patch at a location output by a local estimation approach is replaced by real orientation patch in the local dictionary at the same location. The precondition of applying localized dictionaries is that the pose of the latent fingerprint needs to be estimated. We propose a Hough transform-based fingerprint pose estimation algorithm, in which the predictions about fingerprint pose made by all orientation patches in the latent fingerprint are accumulated. Experimental results on challenging latent fingerprint datasets show the proposed method outperforms previous ones markedly.

In recent years, law enforcement agencies are increasingly using palmprint to identify criminals. For law enforcement palmprint identification systems, efficiency is a very important but challenging problem because of large database size and poor image quality. Existing palmprint identification systems are not sufficiently fast for practical applications. To solve this problem, a novel palmprint indexing algorithm based on ridge features is proposed in this paper. A palmprint is pre-aligned by registering its orientation field with respect to a set of reference orientation fields, which are obtained by clustering training palmprint orientation fields. Indexing is based on comparing ridge orientation fields and ridge density maps, which is much faster than minutiae matching. Proposed algorithm achieved an error rate of 1% at a penetration rate of 2.25% on a palm print database consisting of 13,416 palmprints. Searching a query palmprint over the whole database takes only 0.22 seconds.