Small fiber neuropathy analysis in the epidermis of the skin is critical to the diagnosis of various diseases such as diabetes mellitus and familial amyloid polyneuropathy. Intraepidermal nerve fiber examination plays an important role to serve as a highly sensitive indicator of sensory nerve fiber damage. Segmentation of the epidermis region is a prior image processing task for subsequent image analysis. Due to the lack of suitable and available tools for this specific task, this paper investigates an automatic epidermis segmentation framework in skin fluorescence images. To maintain the high resolution and increase the image amount, the original image is decomposed into a series of sub-images for model input. A U-shaped network structure consisting of five block layers with different depths is proposed for epidermis segmentation. Each model block is designed in a multi-scale manner, which is composed of two different convolution kernel pipelines. By combining and sharing the information from these model blocks with skip connections between them, we can effectively increase the diversity of the feature maps for more accurate segmentation. Experiments on in-house epidermis images demonstrated the advantages of the proposed epidermis segmentation network, which outperformed many state-of-the-art deep learning-based segmentation models. The proposed network architecture is potential in facilitating the epidermis segmentation task for further small fiber neuropathy research.

Accurate registration of remote sensing images through automatic pipelines remains challenging. While bottlenecks have deferred the advancement of traditional approaches, more attention has been attracted on the incorporation of deep learning knowledge into the image registration process. This paper develops an efficient remote sensing image registration framework based upon a convolutional neural network (CNN) architecture, which is called the geometric correlation regression with dense feature network (GcrDfNet). To acquire deep features of remote sensing images, the DenseNet associated with partial transfer learning and partial parameter fine-tuning is exploited. The feature maps derived from the sensed and reference images are further analyzed using a geometric matching model followed by linear regression to compute their correlation and to estimate the transformation coefficients. Subsequently, a network pruning scheme is investigated to diminish the model structure while moderately escalating the registration accuracy. A wide variety of multitemporal and multispectral remote sensing images with distinctive scenarios were employed to evaluate the proposed image registration system. The ensemble parameter compression ratio was approximately 2.12 while slightly reducing the registration error. Experimental results indicated that our GcrDfNet outperformed the traditional and deep learning-based state-of-the-art methods both qualitatively and quantitatively. It is believed that this new image registration model is promising in many remote sensing image processing and analysis applications.

Automatic registration of remote sensing images is an essential task that requires the establishment of appropriate correspondences between the sensed image and the reference image. Among the feature-based matching approaches, attempts relied on the scale-invariant feature transform (SIFT) algorithm have shown particular superiority over other methods. From the perspective of automatic registration, the challenges of SIFT-based methods involve the elimination of mismatches and the homogeneity of matches. While many outlier removal methods have been proposed, the strategies for uniform distribution have been lacking. To address these two issues, this article investigates a new remote sensing image registration algorithm based upon a revised SIFT scheme. Additionally, an outlier removal mechanism founded on a trilateral computation (Tc) recipe and a homogeneity enforcement (He) layout according to a divide-and-conquer inclusion tactic are proposed. Finally, a stochastic competition process based upon game theory is introduced to secure an appropriate amount of correct matches in the proposed Tc and He (TcHe)-SIFT framework. A wide variety of multispectral and multitemporal remote sensing images with various scenarios were exploited to evaluate the proposed registration algorithm. Experimental results demonstrated the advantages of our developed remote sensing image registration algorithm over the state-of-the-art SIFT-based methods. We believe that this new registration technique is of potential in a number of remote sensing image processing applications.

In feature-based remote sensing image registration, the scale-invariant feature transform (SIFT) algorithm has been one of the most popular solutions. However, it is still a challenge to possess an appropriate amount of correct matches while eliminating mismatches. In this letter, inspired by SIFT, an accurate and robust feature matching framework based on feature slope grouping (FSG) for remote sensing image registration is proposed. Our FSG-SIFT algorithm consists of four major phases: modified SIFT, feature slope computation, feature point grouping, and outlier removal and transformation. Specifically, the random sample consensus is adopted to refine the matches followed by the affine transform. The proposed remote sensing image registration algorithm has been validated on a wide variety of high-resolution orthoimagery data. Experimental results with multispectral and multitemporal images suggested that this new image registration algorithm well improved the feature matching accuracy with better registration performance over five state-of-the-art methods.

Vulnerability of recognition systems to spoofing attacks (presentation attacks) is still an open security issue in the biometrics domain. Among all biometric traits, face is exposed to the most serious threat since it is particularly easy to access and reproduce. In this paper, an effective approach against face spoofing attacks based on perceptual image quality assessment features with multiscale analysis is presented. First, we demonstrate that the recently proposed blind image quality evaluator (BIQE) is effective in detecting spoofing attacks. Next, we combine the BIQE with an image quality assessment model called effective pixel similarity deviation (EPSD), which we propose to obtain the standard deviation of the gradient magnitude similarity map by selecting effective pixels in the image. A total number of 21 features acquired from the BIQE and EPSD constitute the multi-scale descriptor for classification. Extensive experiments based on both intradataset and cross-dataset protocols were performed using three existing benchmarks, namely, Replay-Attack, CASIA, and UVAD. The proposed algorithm demonstrated its superiority in detecting face spoofing attacks over many state of the art methods. We believe that the incorporation of the image quality assessment knowledge into face liveness detection is promising to improve the overall accuracy.

Physics-based particle systems are an effective tool for shape modeling. Also, there has been much interest in the study of shape modeling using deformable contour approaches. In this paper, we describe a new deformable model with electric flows based upon computer simulations of a number of charged particles embedded in an electrostatic system. Making use of optimized numerical techniques, the electric potential associated with the electric field in the simulated system is rapidly calculated using the finite-size particle (FSP) method. The simulation of deformation evolves based upon the vector sum of two interacting forces: one from the electric fields and the other from the image gradients. Inspired by the concept of the signed distance function associated with the entropy condition in the level set framework, we efficiently handle topological changes at the interface. In addition to automatic splitting and merging, the evolving contours enable simultaneous detection of various objects with varying intensity gradients at both interior and exterior boundaries. This electric flows approach for shape modeling allows one to connect electric properties in electrostatic equilibrium and classical active contours based upon the theory of curve evolution. Our active contours can be applied to model arbitrarily complicated objects including shapes with sharp corners and cusps, and to situations where no a priori knowledge about the object's topology and geometry is made. We demonstrate the capabilities of this new algorithm in recovering a wide variety of structures on simulated and real images in both 2D and 3D.