A5009982931- Face recognition and analysis
- Face and Expression Recognition
- Biometric Identification and Security
- Video Surveillance and Tracking Methods
- Speech and Audio Processing
- Advanced Image and Video Retrieval Techniques
- Generative Adversarial Networks and Image Synthesis
- Video Analysis and Summarization
- Emotion and Mood Recognition
- Image Retrieval and Classification Techniques
- Advanced Image Processing Techniques
- Advanced Neural Network Applications
- Human Pose and Action Recognition
- Advanced Vision and Imaging
- Gait Recognition and Analysis
- Hand Gesture Recognition Systems
- Handwritten Text Recognition Techniques
- Speech Recognition and Synthesis
- Anomaly Detection Techniques and Applications
- Music and Audio Processing
- Image and Signal Denoising Methods
- Image Enhancement Techniques
- Digital Media Forensic Detection
- Image Processing Techniques and Applications
- Indoor and Outdoor Localization Technologies
Istanbul Technical University
2016-2025
New York University
2025
Qatar University
2024
Stantec (Canada)
2022
École Polytechnique Fédérale de Lausanne
2017-2021
Istanbul University
2016
Karlsruhe Institute of Technology
2005-2013
Boğaziçi University
2004-2012
Karlsruhe University of Education
2005-2007
Sabancı Üniversitesi
2004-2005
In this paper, we present an end-to-end network, called Cycle-Dehaze, for single image dehazing problem, which does not require pairs of hazy and corresponding ground truth images training. That is, train the network by feeding clean in unpaired manner. Moreover, proposed approach rely on estimation atmospheric scattering model parameters. Our method enhances CycleGAN formulation combining cycle-consistency perceptual losses order to improve quality textural information recovery generate...
We present a new dataset for form understanding in noisy scanned documents (FUNSD) that aims at extracting and structuring the textual content of forms. The comprises 199 real, fully annotated, are vary widely appearance, making (FoUn) challenging task. proposed can be used various tasks, including text detection, optical character recognition, spatial layout analysis, entity labeling/linking. To best our knowledge, this is first publicly available with comprehensive annotations to address...
By benefiting from perceptual losses, recent studies have improved significantly the performance of superresolution task, where a high-resolution image is resolved its low-resolution counterpart. Although such objective functions generate near-photorealistic results, their capability limited, since they estimate reconstruction error for an entire in same way, without considering any semantic information. In this paper, we propose novel method to benefit loss more way. We optimize deep...
Deep learning based approaches have been dominating the face recognition field due to significant performance improvement they provided on challenging wild datasets. These extensively tested such unconstrained datasets, Labeled Faces in Wild and YouTube Faces, name a few. However, their capability handle individual appearance variations caused by factors as head pose, illumination, occlusion, misalignment has not thoroughly assessed till now. In this paper, we present comprehensive study...
Face recognition approaches that are based on deep convolutional neural networks (CNN) have been dominating the field. The performance improvements they provided in so called in-the-wild datasets significant, however, their under image quality degradations not assessed, yet. This is particularly important, since real-world face applications, images may contain various kinds of due to motion blur, noise, compression artifacts, color distortions, and occlusion. In this work, we addressed...
This paper reviews the first challenge on image dehazing (restoration of rich details in hazy image) with focus proposed solutions and results. The had 2 tracks. Track 1 employed indoor images (using I-HAZE dataset), while outdoor O-HAZE dataset). have been captured presence real haze, generated by professional haze machines. dataset contains 35 scenes that correspond to domestic environments, objects different colors specularities. 45 depicting same visual content recorded haze-free...
Convolutional neural network (CNN) based approaches are the state of art in various computer vision tasks including face recognition. Considerable research effort is currently being directed toward further improving CNNs by focusing on model architectures and training techniques. However, studies systematically exploring strengths weaknesses existing deep models for recognition still relatively scarce. In this paper, we try to fill gap study effects different covariates verification...
Vision-based action recognition is one of the most challenging research topics computer vision and pattern recognition. A specific application it, namely, detecting fights from surveillance cameras in public areas, prisons, etc., desired to quickly get under control these violent incidents. This paper addresses this problem explores LSTM-based approaches solve it. Moreover, attention layer also utilized. Besides, a new dataset collected, which consists fight scenes camera videos available at...
Age and gender are complementary soft biometric traits for face recognition. Successful estimation of age from facial images taken under real-world conditions can contribute improving the identification results in wild. In this study, order to achieve robust classification wild, we have benefited Deep Convolutional Neural Networks based representation. We explored transferability existing deep convolutional neural network (CNN) models classification. The generic AlexNet-like architecture...
In this paper, we present our work in building technologies for natural multimodal human-robot interaction. We systems spontaneous speech recognition, dialogue processing, and visual perception of a user, which includes localization, tracking, identification the recognition pointing gestures, as well person's head orientation. Each components is described paper experimental results are presented. also several experiments on interaction, such interaction using automatic determination...
In this paper, we present the classification sub-system of a real-time video-based face identification system which recognizes people entering through door laboratory. Since subjects are not asked to cooperate with but allowed behave naturally, application scenario poses many challenges. Continuous, uncontrolled variations facial appearance due illumination, pose, expression, and occlusion need be handled allow for successful recognition. Faces classified by local appearance-based...
In this paper, we have extensively investigated the unconstrained ear recognition problem. We first shown importance of domain adaptation, when deep convolutional neural network models are used for recognition. To enable collected a new dataset using Multi-PIE face dataset, which named as dataset. improve performance further, combined different models. analyzed in depth effect image quality, example illumination and aspect ratio, on classification performance. Finally, addressed problem bias...
In this paper we present the results of Unconstrained Ear Recognition Challenge (UERC), a group benchmarking effort centered around problem person recognition from ear images captured in uncontrolled conditions. The goal challenge was to assess performance existing techniques on challenging large-scale dataset and identify open problems that need be addressed future. Five groups three continents participated contributed six for evaluation, while multiple baselines were made available by UERC...
In this paper, we address the problem of apparent age estimation. Different from estimating real individuals, in which each face image has a single label, problem, images have multiple labels, corresponding to ages perceived by annotators, when they look at these images. This provides an intriguing computer vision since generic or object classification tasks, it is typical ground truth label per class. To account for labels image, instead using average annotated as class grouped that are...
This paper presents a summary of the Masked Face Recognition Competitions (MFR) held within 2021 International Joint Conference on Biometrics (IJCB 2021). The competition attracted total 10 participating teams with valid submissions. affiliations these are diverse and associated academia industry in nine different countries. These successfully submitted 18 solutions. is designed to motivate solutions aiming at enhancing face recognition accuracy masked faces. Moreover, considered...
Anomalies are ubiquitous in all scientific fields and can express an unexpected event due to incomplete knowledge about the data distribution or unknown process that suddenly comes into play distorts observations. Usually, such events' rarity, train deep learning (DL) models on anomaly detection (AD) task, scientists only rely "normal" data, i.e., nonanomalous samples. Thus, letting neural network infer beneath input data. In a context, we propose novel framework, named multilayer one-class...
In this paper, the effects of feature selection and normalization to performance a local appearance based face recognition scheme are presented. From features that extracted using block-based discrete cosine transform, three sets derived. These vectors normalized in two different ways; by making them unit norm dividing each coefficient its standard deviation is learned from training set. The input test images then classified four distance measures: L1 norm, L2 angle covariance between...
For the last 50 years, intelligent tutoring systems have been developed with aim to supporting one of most successful educational forms – individual teaching. Recent research has shown that emotions can influence human behavior and learning abilities, as a result developers also started follow these ideas by creating affective systems. However, adaptation skills mentioned type are still imperfect. The paper presents an analysis emotion recognition methods used in existing enhance ongoing on...
Health organizations advise social distancing, wearing face mask, and avoiding touching to prevent the spread of coronavirus. Based on these protective measures, we developed a computer vision system help transmission COVID-19. Specifically, performs mask detection, face-hand interaction measures distance. To train evaluate system, collected annotated images that represent usage in real world. Besides assessing performance our own datasets, also tested it existing datasets literature without...
Recent state-of-the-art methods for skin lesion segmentation are based on convolutional neural networks (CNNs). Even though these CNN-based approaches accurate, they computationally expensive. In this paper, we address problem and propose an efficient fully network, named DermoNet. DermoNet, due to our densely connected blocks skip connections, network layers can reuse information from their preceding ensure high accuracy in later layers. By doing so, take advantage of the capability...
Pedestrian detection is an important component for safety of autonomous vehicles, as well traffic and street surveillance. There are extensive benchmarks on this topic it has been shown to be a challenging problem when applied real use-case scenarios. In purely image-based pedestrian approaches, the state-of-the-art results have achieved with convolutional neural networks (CNN) surprisingly few frameworks built upon multi-cue approaches. work, we develop new detector vehicles that exploits...