A5028776484- Digital Media Forensic Detection
- Face recognition and analysis
- Adversarial Robustness in Machine Learning
- Generative Adversarial Networks and Image Synthesis
- Domain Adaptation and Few-Shot Learning
- Biometric Identification and Security
- Face and Expression Recognition
- Anomaly Detection Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Cell Image Analysis Techniques
- Image Retrieval and Classification Techniques
- Handwritten Text Recognition Techniques
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Image Processing Techniques and Applications
- AI in cancer detection
- Video Surveillance and Tracking Methods
- Advanced Vision and Imaging
- Forensic Fingerprint Detection Methods
- Topic Modeling
- Vehicle License Plate Recognition
- Misinformation and Its Impacts
- Advanced Malware Detection Techniques
- Neural Networks and Applications
Clemson University
2024-2025
Marina Del Rey Hospital
2021-2024
University of Southern California
2015-2022
LAC+USC Medical Center
2022
Southern California University for Professional Studies
2019-2021
Loyola Marymount University
2018
Norwegian University of Science and Technology
2018
National University of Singapore
2018
University of Maryland, College Park
2005-2014
EarthTech International (United States)
2008
To fight against real-life image forgery, which commonly involves different types and combined manipulations, we propose a unified deep neural architecture called ManTraNet. Unlike many existing solutions, ManTra-Net is an end-to-end network that performs both detection localization without extra preprocessing postprocessing. fully convolutional handles images of arbitrary sizes known forgery such splicing, copy-move, removal, enhancement, even unknown types. This paper has three salient...
The spread of misinformation through synthetically generated yet realistic images and videos has become a significant problem, calling for robust manipulation detection methods. Despite the predominant effort detecting face in still images, less attention been paid to identification tampered faces by taking advantage temporal information present stream. Recurrent convolutional models are class deep learning which have proven effective at exploiting from image streams across domains. We...
We introduce our method and system for face recognition using multiple pose-aware deep learning models. In representation, a image is processed by several pose-specific convolutional neural network (CNN) models to generate features. 3D rendering used poses from the input image. Sensitivity of pose variations reduced since we use an ensemble CNN The paper presents extensive experimental results on effect landmark detection, layer selection model performance pipeline. Our novel representation...
We propose a method designed to push the frontiers of unconstrained face recognition in wild with an emphasis on extreme out-of-plane pose variations. Existing methods either expect single model learn invariance by training massive amounts data or else normalize images aligning faces frontal pose. Contrary these, our is explicitly tackle Our proposed Pose-Aware Models (PAM) process image using several pose-specific, deep convolutional neural networks (CNN). 3D rendering used synthesize...
In this paper, for the first time, we introduce a new end-to-end deep neural network predicting forgery masks to image copy-move detection problem. Specifically, use convolutional extract block-like features from an image, compute self-correlations between different blocks, pointwise feature extractor locate matching points, and reconstruct mask through deconvolutional network. Unlike classic solutions requiring multiple stages of training parameter tuning, ranging extraction postprocessing,...
Image splicing is a very common image manipulation technique that sometimes used for malicious purposes. A detection and localization algorithm usually takes an input produces binary decision indicating whether the has been manipulated, also segmentation mask corresponds to spliced region. Most existing pipelines suffer from two main shortcomings: 1) they use handcrafted features are not robust against subsequent processing (e.g., compression), 2) each stage of pipeline optimized...
Recently, generative adversarial networks and autoencoders have gained a lot of attention in machine learning community due to their exceptional performance tasks such as digit classification face recognition.They map the autoencoder's bottleneck layer output (termed code vectors) different noise Probability Distribution Functions (PDFs), that can be further regularized cluster based on class information.In addition, they also allow generation synthetic samples by sampling vectors from...
Finding a template in search image is one of the core problems many computer vision applications, such as matching, semantic alignment, image-to-GPS verification \etc. In this paper, we propose novel quality-aware matching method, which not only used standalone algorithm, but also trainable layer that can be easily plugged any deep neural network. Specifically, assess quality pair its soft-ranking among all pairs, and thus different scenarios like 1-to-1, 1-to-many, many-to-many will...
Few-shot Learning has been studied to mimic human visual capabilities and learn effective models without the need of exhaustive annotation. Even though idea meta-learning for adaptation dominated few-shot learning methods, how train a feature extractor is still challenge. In this paper, we focus on design training strategy obtain an elemental representation such that prototype each novel class can be estimated from few labeled samples. We propose two-stage scheme, Partner-Assisted (PAL),...
The vulnerability of biometric identification systems to presentation attacks, also a.k.a. spoof has received great attention from the biometrics community. Face attacks are particularly easy fabricate, because face image and videos obtain social media, most recognition rely on 2D RGB sensors only. To our best knowledge, all existing Presentation Attack Detection (PAD) research works focus solving problem in a close-set setup, even though it is an open-set real life where Unknown (UPAD)...
Unlike conventional zero-shot classification, semantic segmentation predicts a class label at the pixel level instead of image level. When solving problems, need for pixel-level prediction with surrounding context motivates us to incorporate spatial information using positional encoding. We improve standard encoding by introducing concept Relative Positional Encoding, which integrates feature and can handle arbitrary sizes. Furthermore, while self-training is widely used in generate...
In this paper, we present a novel algorithm for simultaneously detecting shot boundaries and extracting key frames from video sequences or streams in real-time. Multivariate feature vectors are extracted the arranged matrix. Singular value decomposition is then used to factorize matrix compute significant singular vectors. The rank of traced using sliding window approach. By tracing computed rank, able determine extract video. Results experiments conducted on soccer videos show that robust...
In this paper, we present a novel graph-based method for extracting handwritten text lines in monochromatic Arabic document images. Our approach consists of two steps - Coarse line estimation using primary components which define the and assignment diacritic are more difficult to associate with given line. We first estimate local orientation at each component build sparse similarity graph. then, use shortest path algorithm compute similarities between non-neighboring components. From graph,...
We propose a novel approach to achieving invariance for deep neural networks in the form of inducing amnesia unwanted factors data through new adversarial forgetting mechanism. show that mechanism serves as an information-bottleneck, which is manipulated by training learn factors. Empirical results proposed framework achieves state-of-the-art performance at learning both nuisance and bias settings on diverse collection datasets tasks.
Advances in deep learning, combined with availability of large datasets, have led to impressive improvements face presentation attack detection research. However, state-of-the-art antispoofing systems are still vulnerable novel types attacks that never seen during training. Moreover, even if such correctly detected, these lack the ability adapt newly encountered attacks. The post-training continually detecting new and self-adaptation identify types, after initial phase, is highly appealing....
We introduce a novel capture device that provides two new powerful sensing modalities for fingerprint presentation attack detection (FPAD): multi-spectral short-wave infrared (SWIR) imaging and laser speckle contrast (LSCI). A prototype touchless was developed employing 64×64 indium gallium arsenide (InGaAs) sensor high definition visible light camera. The former capabilities while the latter enables backward compatibility with legacy matching systems. patch-based convolutional neural...
This paper presents an approach to text line extraction in handwritten document images which combines local and global techniques. We propose a graph-based technique detect touching proximity errors that are common with lines. In refinement step, we use Expectation-Maximization (EM) iteratively split the error segments obtain correct text-lines. show improvement accuracies using our correction method on datasets of Arabic images. Results set artificially generated is effective for handling
We present a fully trainable solution for binarization of degraded document images using extremely randomized trees. Unlike previous attempts that often use simple features, our method encodes all heuristics about whether or not pixel is foreground text into high-dimensional feature vector and learns more complicated decision function. introduce two novel the Logarithm Intensity Percentile (LIP) Relative Darkness Index (RDI), combine them with low level reformulated features from existing...