Lei Li’s research while affiliated with Shandong University of Finance and Economics and other places

... To identify spoofing attack by combining the result of two separate technique i.e head 3D properties along with the eye blink. In the literature, for joint operation of face verification and face anti-spoofing [7], The two fusion techniques decision and score level were adopted to combine the face verification and anti-spoofing. In order to test the fusion, four rules have been set for fusion strategies: AND, SUM, Logistic Regression (LR) and Polynomial Logistic Regression (PLR). ...

Reference:

Robust Hybrid Texture Descriptor (HTD) and a parallel score based fusion for face verification and liveness detection system

... Furthermore, the advent of deep learning (DL) has elicited favorable responses in crack detection research, with DL object detection technologies gradually permeating the domain of non-destructive testing within wooden structures [4][5][6][7][8]. Present research predominantly harnesses convolutional neural networks and YOLO series models to discern and categorize wooden architectural sites [9,10], wood, and wooden utensils [11,12] by extracting texture features and analysis of color attributes. ...

Reference:

ICDW-YOLO: An Efficient Timber Construction Crack Detection Algorithm

... However, the bottom layer of the pyramid structure does not effectively capture the semantic information of the higher layers, and methods that use a U-shaped structure to enable the fusion of the higher layer with the bottom layer features through convolutional and anti-convolutional layers have been developed. The two-stream-based I3D network can capture both temporal and spatial information [24] and the architecture combined with transformer can further capture relationships [25].In addition to adjusting the structure of the classification network, there are also ways to reduce label noise to improve the model's effectiveness [26]. The development of fully supervised approaches is limited because they require a large amount of manual annotation work and do not clearly define the start and end times of actions. ...

Reference:

TSCANet: a two-stream context aggregation network for weakly-supervised temporal action localization

... Motivated by the success of deep learning methods in image recognition, deep neural networks have been extensively explored for automatically learning the features from facial expression images [16][17][18]. Building on this foundation, they have been applied to pain assessment [19][20][21]. The convolutional neural networks (CNNs), e.g., VGG-16 pre-trained with millions of face images, were used to learn the facial features of image flame for pain estimation, which can alleviate the problem of less training data in facial expression of pain [19]. ...

Reference:

Hierarchical global and local transformer for pain estimation with facial expression videos

... Several defense mechanisms have been proposed for PAI attacks. Optical flow texture analysis, for instance, detects subtle motion inconsistencies in spoofed biometric data by tracking pixel-level movements, effectively distinguishing real subjects from static or fake presentation attacks [Li et al. [2022], Bhattacharjee et al. [2019], Damer et al. [2016], Raghavendra et al. [2015], Ramachandra and Busch [2017b]]. Attacks on speech involve generating deepfakes or replaying audio to deceive authentication systems. ...

Reference:

Biometrics in Extended Reality: A Review

... are characterized by limited fusion rules and the high complexity of manual design, have constrained the potential for enhancement in method performance. Although deep learning networks are capable of extracting features with remarkable efficacy, deep learning-based methods [8][9][10][11] have become the prevailing paradigm for image fusion. These methods can be classified into three principal categories [12]: self-encoder-based [13,14], convolutional neural network-based [15,16], and generative adversarial network-based [17][18][19]. ...

Reference:

Multispectral image fusion technique based on decoupling of information

... The model was trained using the triplet loss function to reduce intra-class distance and maximize inter-class distance. The system achieved an HTER of 0.7 [42]. A research study utilized a multi-scale retinex model to separate an image's light and reflectance components. ...

Reference:

Lightweight 3D-StudentNet for defending against face replay attacks

... Deep learning and Convolutional Neural Network (CNN) have achieved great success in many computer vision tasks; however, they suffer from the overfitting problem for FAS tasks due to the limited amount and diversity of the training data. Therefore, the hybrid FAS methods (handcraft + deep learning) were proposed [16][17][18][19][20]. There are three types of hybrid frameworks: extracting deep convolutional features from the handcrafted features [16,17], extracting handcrafted features from deep convolutional features [18], and fusing the handcrafted and deep convolutional features for more generic representation [19]. ...

Reference:

A MobileFaceNet-Based Face Anti-Spoofing Algorithm for Low-Quality Images

... Additionally, in Patel et al. (2016), image distortions were examined using various intensity channels (R, G, B, and grayscale) and across different image regions (entire image, detected face, and the facial area between the nose and chin). In Li and Feng (2019), traditional handcrafted features were combined with convolutional neural networks (CNNs) to enhance face PAD. Lately, a hybrid technique was presented in , combining information on appearance from two CNNs, with an SVM classifier employed to distinguish between live and spoofed images. ...

Reference:

Domain Generalization via Ensemble Stacking for Face Presentation Attack Detection

... Among pain-related research, most efforts have focused on recognition and classification using either traditional machine learning or deep learning techniques [14], [15], [16], [17]. PainDiffusion is the first method to explore pain expression generation using diffusion models, supporting arbitrarylength predictions. ...

Reference:

PainDiffusion: Can robot express pain?