Man Zhang

Conventional iris recognition requires controlled conditions (e.g., close acquisition distance and stop-and-stare scheme) and high user cooperation for image acquisition. Non-cooperative acquisition environments introduce many adverse factors such as blur, off-axis, occlusions and specular reflections, which challenge existing iris segmentation approaches. In this paper, we present two iris segmentation models, namely hierarchical convolutional neural networks (HCNNs) and multi-scale fully convolutional network (MFCNs), for noisy iris images acquired at-a-distance and on-the-move. Both models automatically locate iris pixels without handcrafted features or rules. Moreover, the features and classifiers are jointly optimized. They are end-to-end models which require no further pre- and post-processing and outperform other state-of-the-art methods. Compared with HCNNs, MFCNs take input of arbitrary size and produces correspondingly-sized output without sliding window prediction, which makes MFCNs more efficient. The shallow, fine layers and deep, global layers are combined in MFCNs to capture both the texture details and global structure of iris patterns. Experimental results show that MFCNs are more robust than HCNNs to noises, and can greatly improve the current state-of-the-arts by 25.62% and 13.24% on the UBIRIS.v2 and CASIA.v4-distance databases, respectively.

Emission of heavy metals from traffic activities is an important pollution source to roadside farmland ecosystems. However, little previous research has been conducted to investigate heavy metal concentrations of roadside farmland soil in mountainous areas. Owing to more complex roadside environments and more intense driving conditions on mountainous highways, heavy metal accumulation and distribution patterns in farmland soil due to traffic activity could be different from those on plain highways. In this study, design factors including altitude, roadside distance, terrain, and tree protection were considered to analyze their influences on Cu, Zn, Cd, and Pb concentrations in farmland soils along a mountain highway around Kathmandu, Nepal. On average, the concentrations of Cu, Zn, Cd, and Pb at the sampling sites are lower than the tolerable levels. Correspondingly, pollution index analysis does not show serious roadside pollution owing to traffic emissions either. However, some maximum Zn, Cd, and Pb concentrations are close to or higher than the tolerable level, indicating that although average accumulations of heavy metals pose no hazard in the region, some spots with peak concentrations may be severely polluted. The correlation analysis indicates that either Cu or Cd content is found to be significantly correlated with Zn and Pb content while there is no significant correlation between Cu and Cd. The pattern can be reasonably explained by the vehicular heavy metal emission mechanisms, which proves the heavy metals' homology of the traffic pollution source. Furthermore, the independent factors show complex interaction effects on heavy metal concentrations in the mountainous roadside soil, which indicate quite a different distribution pattern from previous studies focusing on urban roadside environments. It is found that the Pb concentration in the downgrade roadside soil is significantly lower than that in the upgrade soil while the Zn concentration in the downgrade roadside soil is marginally higher than in the upgrade soil; and the concentrations of Cu and Pb in the roadside soils with tree protection are significantly lower than those without tree protection. However, the attenuation pattern of heavy metal concentrations as a function of roadside distance within a 100 m range cannot be identified consistently.

The gap between sensing patterns of different face modalities remains a challenging problem in heterogeneous face recognition (HFR). This paper proposes an adversarial discriminative feature learning framework to close the sensing gap via adversarial learning on both raw-pixel space and compact feature space. This framework integrates cross-spectral face hallucination and discriminative feature learning into an end-to-end adversarial network. In the pixel space, we make use of generative adversarial networks to perform cross-spectral face hallucination. An elaborate two-path model is introduced to alleviate the lack of paired images, which gives consideration to both global structures and local textures. In the feature space, an adversarial loss and a high-order variance discrepancy loss are employed to measure the global and local discrepancy between two heterogeneous distributions respectively. These two losses enhance domain-invariant feature learning and modality independent noise removing. Experimental results on three NIR-VIS databases show that our proposed approach outperforms state-of-the-art HFR methods, without requiring of complex network or large-scale training dataset.