Publications

@article{GrmAccess2025,

title = {Degrade or super-resolve to recognize? Bridging the Domain Gap for Cross-Resolution Face Recognition},

author = {Klemen Grm and Berk Kemal Ozata and Alperen Kantarci and Vitomir Struc and Hazim Kemal Ekenel},

url = {https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10833634

https://lmi.fe.uni-lj.si/wp-content/uploads/2025/01/Degrade_or_super-resolve_to_recognize_Bridging_the_Domain_Gap_for_Cross-Resolution_Face_Recognition_compressed-1.pdf},

doi = {10.1109/ACCESS.2025.3527236},

issn = {2169-3536},

year  = {2025},

date = {2025-01-08},

journal = {IEEE Access},

pages = {1-16},

abstract = {In this work, we address the problem of cross-resolution face recognition, where a low-resolution probe face is compared against high-resolution gallery faces. To address this challenging problem, we investigate two approaches for bridging the quality gap between low-quality probe faces and high-quality gallery faces. The first approach focuses on degrading the quality of high-resolution gallery images to bring them closer to the quality of the probe images. The second approach involves enhancing the resolution of the probe images using face hallucination. Our experiments on the SCFace and DroneSURF datasets reveal that the success of face hallucination is highly dependent on the quality of the original images, since poor image quality can severely limit the effectiveness of the hallucination technique. Therefore, the selection of the appropriate face recognition method should consider the quality of the images. Additionally, our experiments also suggest that combining gallery degradation and face hallucination in a hybrid recognition scheme provides the best overall results for cross-resolution face recognition with relatively high-quality probe images, while the degradation process on its own is the more suitable option for low-quality probe images. Our results show that the combination of standard computer vision approaches such as degradation, super-resolution, feature fusion, and score fusion can be used to substantially improve performance on the task of low resolution face recognition using off-the-shelf face recognition models without re-training on the target domain.},

keywords = {CNN, face recognition, low quality, super-resolution},

pubstate = {published},

tppubtype = {article}

}

@article{Vitek2024_Gaze,

title = {GazeNet: A lightweight multitask sclera feature extractor},

author = {Matej Vitek and Vitomir Štruc and Peter Peer},

url = {https://lmi.fe.uni-lj.si/wp-content/uploads/2024/11/1-s2.0-S1110016824014273-main.pdf

https://www.sciencedirect.com/science/article/pii/S1110016824014273},

doi = {https://doi.org/10.1016/j.aej.2024.11.011},

year  = {2025},

date = {2025-01-05},

journal = {Alexandria Engineering Journal},

volume = {112},

pages = {661-671},

abstract = {The sclera is a recently emergent biometric modality with many desirable characteristics. However, most literature solutions for sclera-based recognition rely on sequences of complex deep networks with significant computational overhead. In this paper, we propose a lightweight multitask-based sclera feature extractor. The proposed GazeNet network has a computational complexity below 1 GFLOP, making it appropriate for less capable devices like smartphones and head-mounted displays. Our experiments show that GazeNet (which is based on the SqueezeNet architecture) outperforms both the base SqueezeNet model as well as the more computationally intensive ScleraNET model from the literature. Thus, we demonstrate that our proposed gaze-direction multitask learning procedure, along with careful lightweight architecture selection, leads to computationally efficient networks with high recognition performance.},

keywords = {biometrics, CNN, deep learning, lightweight models, sclera},

pubstate = {published},

tppubtype = {article}

}

@article{Darian2024,

title = {Generating bimodal privacy-preserving data for face recognition},

author = {Darian Tomašević and Fadi Boutros and Naser Damer and Peter Peer and Vitomir Štruc},

url = {https://lmi.fe.uni-lj.si/wp-content/uploads/2024/05/PapersDarian.pdf},

doi = {https://doi.org/10.1016/j.engappai.2024.108495},

year  = {2024},

date = {2024-05-01},

journal = {Engineering Applications of Artificial Intelligence},

volume = {133},

issue = {E},

pages = {1-25},

abstract = {The performance of state-of-the-art face recognition systems depends crucially on the availability of large-scale training datasets. However, increasing privacy concerns nowadays accompany the collection and distribution of biometric data, which has already resulted in the retraction of valuable face recognition datasets. The use of synthetic data represents a potential solution, however, the generation of privacy-preserving facial images useful for training recognition models is still an open problem. Generative methods also remain bound to the visible spectrum, despite the benefits that multispectral data can provide. To address these issues, we present a novel identity-conditioned generative framework capable of producing large-scale recognition datasets of visible and near-infrared privacy-preserving face images. The framework relies on a novel identity-conditioned dual-branch style-based generative adversarial network to enable the synthesis of aligned high-quality samples of identities determined by features of a pretrained recognition model. In addition, the framework incorporates a novel filter to prevent samples of privacy-breaching identities from reaching the generated datasets and improve both identity separability and intra-identity diversity. Extensive experiments on six publicly available datasets reveal that our framework achieves competitive synthesis capabilities while preserving the privacy of real-world subjects. The synthesized datasets also facilitate training more powerful recognition models than datasets generated by competing methods or even small-scale real-world datasets. Employing both visible and near-infrared data for training also results in higher recognition accuracy on real-world visible spectrum benchmarks. Therefore, training with multispectral data could potentially improve existing recognition systems that utilize only the visible spectrum, without the need for additional sensors.},

keywords = {CNN, face, face generation, face images, face recognition, generative AI, StyleGAN2, synthetic data},

pubstate = {published},

tppubtype = {article}

}

@article{BabnikTBIOM2024,

title = {eDifFIQA: Towards Efficient Face Image Quality Assessment based on Denoising Diffusion Probabilistic Models},

author = {Žiga Babnik and Peter Peer and Vitomir Štruc},

url = {https://lmi.fe.uni-lj.si/wp-content/uploads/2024/03/TBIOM___DifFIQAv2.pdf

https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10468647&tag=1},

doi = {10.1109/TBIOM.2024.3376236},

issn = {2637-6407},

year  = {2024},

date = {2024-03-07},

urldate = {2024-03-07},

journal = {IEEE Transactions on Biometrics, Behavior, and Identity Science (TBIOM)},

pages = {1-16},

abstract = {State-of-the-art Face Recognition (FR) models perform well in constrained scenarios, but frequently fail in difficult real-world scenarios, when no quality guarantees can be made for face samples. For this reason, Face Image Quality Assessment (FIQA) techniques are often used by FR systems, to provide quality estimates of captured face samples. The quality estimate provided by FIQA techniques can be used by the FR system to reject samples of low-quality, in turn improving the performance of the system and reducing the number of critical false-match errors. However, despite steady improvements, ensuring a good trade-off between the performance and computational complexity of FIQA methods across diverse face samples remains challenging. In this paper, we present DifFIQA, a powerful unsupervised approach for quality assessment based on the popular denoising diffusion probabilistic models (DDPMs) and the extended (eDifFIQA) approach. The main idea of the base DifFIQA approach is to utilize the forward and backward processes of DDPMs to perturb facial images and quantify the impact of these perturbations on the corresponding image embeddings for quality prediction. Because of the iterative nature of DDPMs the base DifFIQA approach is extremely computationally expensive. Using eDifFIQA we are able to improve on both the performance and computational complexity of the base DifFIQA approach, by employing label optimized knowledge distillation. In this process, quality information inferred by DifFIQA is distilled into a quality-regression model. During the distillation process, we use an additional source of quality information hidden in the relative position of the embedding to further improve the predictive capabilities of the underlying regression model. By choosing different feature extraction backbone models as the basis for the quality-regression eDifFIQA model, we are able to control the trade-off between the predictive capabilities and computational complexity of the final model. We evaluate three eDifFIQA variants of varying sizes in comprehensive experiments on 7 diverse datasets containing static-images and a separate video-based dataset, with 4 target CNN-based FR models and 2 target Transformer-based FR models and against 10 state-of-the-art FIQA techniques, as well as against the initial DifFIQA baseline and a simple regression-based predictor DifFIQA(R), distilled from DifFIQA without any additional optimization. The results show that the proposed label optimized knowledge distillation improves on the performance and computationally complexity of the base DifFIQA approach, and is able to achieve state-of-the-art performance in several distinct experimental scenarios. Furthermore, we also show that the distilled model can be used directly for face recognition and leads to highly competitive results.},

keywords = {biometrics, CNN, deep learning, DifFIQA, difussion, face, face image quality assesment, face recognition, FIQA},

pubstate = {published},

tppubtype = {article}

}

@article{ESWA2024,

title = {Y-GAN: Learning Dual Data Representations for Anomaly Detection in Images},

author = {Marija Ivanovska and Vitomir Štruc},

url = {https://www.sciencedirect.com/science/article/pii/S0957417424002756

https://lmi.fe.uni-lj.si/wp-content/uploads/2024/02/YGAN_Marija.pdf},

doi = {https://doi.org/10.1016/j.eswa.2024.123410},

year  = {2024},

date = {2024-03-01},

urldate = {2024-03-01},

journal = {Expert Systems with Applications (ESWA)},

volume = {248},

number = {123410},

pages = {1-7},

abstract = {We propose a novel reconstruction-based model for anomaly detection in image data, called 'Y-GAN'. The model consists of a Y-shaped auto-encoder and represents images in two separate latent spaces. The first captures meaningful image semantics, which are key for representing (normal) training data, whereas the second encodes low-level residual image characteristics. To ensure the dual representations encode mutually exclusive information, a disentanglement procedure is designed around a latent (proxy) classifier. Additionally, a novel representation-consistency mechanism is proposed to prevent information leakage between the latent spaces. The model is trained in a one-class learning setting using only normal training data. Due to the separation of semantically-relevant and residual information, Y-GAN is able to derive informative data representations that allow for efficacious anomaly detection across a diverse set of anomaly detection tasks. The model is evaluated in comprehensive experiments with several recent anomaly detection models using four popular image datasets, i.e., MNIST, FMNIST, CIFAR10, and PlantVillage. Experimental results show that Y-GAN outperforms all tested models by a considerable margin and yields state-of-the-art results. The source code for the model is made publicly available at https://github.com/MIvanovska/Y-GAN. },

keywords = {anomaly detection, CNN, deep learning, one-class learning, y-gan},

pubstate = {published},

tppubtype = {article}

}

@article{KrizajEAAI2024,

title = {Deep Face Decoder: Towards understanding the embedding space of convolutional networks through visual reconstruction of deep face templates},

author = {Janez Križaj and Richard O. Plesh and Mahesh Banavar and Stephanie Schuckers and Vitomir Štruc},

url = {https://www.sciencedirect.com/science/article/abs/pii/S095219762400099X

https://lmi.fe.uni-lj.si/wp-content/uploads/2025/02/Deep_Face_Decoder__Elsevier_template_.pdf},

doi = {https://doi.org/10.1016/j.engappai.2024.107941},

year  = {2024},

date = {2024-01-30},

urldate = {2024-01-30},

journal = {Engineering Applications of Artificial Intelligence},

volume = {132},

issue = {107941},

pages = {1-20},

abstract = {Advances in deep learning and convolutional neural networks (ConvNets) have driven remarkable face recognition (FR) progress recently. However, the black-box nature of modern ConvNet-based face recognition models makes it challenging to interpret their decision-making process, to understand the reasoning behind specific success and failure cases, or to predict their responses to unseen data characteristics. It is, therefore, critical to design mechanisms that explain the inner workings of contemporary FR models and offer insight into their behavior. To address this challenge, we present in this paper a novel textit{template-inversion approach} capable of reconstructing high-fidelity face images from the embeddings (templates, feature-space representations) produced by modern FR techniques. Our approach is based on a novel Deep Face Decoder (DFD) trained in a regression setting to visualize the information encoded in the embedding space with the goal of fostering explainability. We utilize the developed DFD model in comprehensive experiments on multiple unconstrained face datasets, namely Visual Geometry Group Face dataset 2 (VGGFace2), Labeled Faces in the Wild (LFW), and Celebrity Faces Attributes Dataset High Quality (CelebA-HQ). Our analysis focuses on the embedding spaces of two distinct face recognition models with backbones based on the Visual Geometry Group 16-layer model (VGG-16) and the 50-layer Residual Network (ResNet-50). The results reveal how information is encoded in the two considered models and how perturbations in image appearance due to rotations, translations, scaling, occlusion, or adversarial attacks, are propagated into the embedding space. Our study offers researchers a deeper comprehension of the underlying mechanisms of ConvNet-based FR models, ultimately promoting advancements in model design and explainability. },

keywords = {CNN, embedding space, face, face images, face recognition, face synthesis, template reconstruction, xai},

pubstate = {published},

tppubtype = {article}

}

@article{MaskFaceGAN,

title = {MaskFaceGAN: High Resolution Face Editing With Masked GAN Latent Code Optimization},

author = {Martin Pernuš and Vitomir Štruc and Simon Dobrišek},

url = {https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10299582

https://lmi.fe.uni-lj.si/wp-content/uploads/2023/02/MaskFaceGAN_compressed.pdf

https://arxiv.org/pdf/2103.11135.pdf},

doi = {10.1109/TIP.2023.3326675},

issn = {1941-0042},

year  = {2023},

date = {2023-10-27},

urldate = {2023-01-02},

journal = {IEEE Transactions on Image Processing},

abstract = {Face editing represents a popular research topic within the computer vision and image processing communities. While significant progress has been made recently in this area, existing solutions: ( i ) are still largely focused on low-resolution images, ( ii ) often generate editing results with visual artefacts, or ( iii ) lack fine-grained control over the editing procedure and alter multiple (entangled) attributes simultaneously, when trying to generate the desired facial semantics. In this paper, we aim to address these issues through a novel editing approach, called MaskFaceGAN that focuses on local attribute editing. The proposed approach is based on an optimization procedure that directly optimizes the latent code of a pre-trained (state-of-the-art) Generative Adversarial Network (i.e., StyleGAN2) with respect to several constraints that ensure: ( i ) preservation of relevant image content, ( ii ) generation of the targeted facial attributes, and ( iii ) spatially–selective treatment of local image regions. The constraints are enforced with the help of an (differentiable) attribute classifier and face parser that provide the necessary reference information for the optimization procedure. MaskFaceGAN is evaluated in extensive experiments on the FRGC, SiblingsDB-HQf, and XM2VTS datasets and in comparison with several state-of-the-art techniques from the literature. Our experimental results show that the proposed approach is able to edit face images with respect to several local facial attributes with unprecedented image quality and at high-resolutions (1024×1024), while exhibiting considerably less problems with attribute entanglement than competing solutions. The source code is publicly available from: https://github.com/MartinPernus/MaskFaceGAN.},

keywords = {CNN, computer vision, deep learning, face editing, face image processing, GAN, GAN inversion, generative models, StyleGAN},

pubstate = {published},

tppubtype = {article}

}

@article{VitekSaud2023,

title = {IPAD: Iterative Pruning with Activation Deviation for Sclera Biometrics},

author = {Matej Vitek and Matic Bizjak and Peter Peer and Vitomir Štruc},

url = {https://lmi.fe.uni-lj.si/wp-content/uploads/2023/07/PublishedVersion.pdf},

doi = {https://doi.org/10.1016/j.jksuci.2023.101630},

year  = {2023},

date = {2023-07-10},

journal = {Journal of King Saud University - Computer and Information Sciences},

volume = {35},

number = {8},

pages = {1-21},

abstract = {The sclera has recently been gaining attention as a biometric modality due to its various desirable characteristics. A key step in any type of ocular biometric recognition, including sclera recognition, is the segmentation of the relevant part(s) of the eye. However, the high computational complexity of the (deep) segmentation models used in this task can limit their applicability on resource-constrained devices such as smartphones or head-mounted displays. As these devices are a common desired target for such biometric systems, lightweight solutions for ocular segmentation are critically needed. To address this issue, this paper introduces IPAD (Iterative Pruning with Activation Deviation), a novel method for developing lightweight convolutional networks, that is based on model pruning. IPAD uses a novel filter-activation-based criterion (ADC) to determine low-importance filters and employs an iterative model pruning procedure to derive the final lightweight model. To evaluate the proposed pruning procedure, we conduct extensive experiments with two diverse segmentation models, over four publicly available datasets (SBVPI, SLD, SMD and MOBIUS), in four distinct problem configurations and in comparison to state-of-the-art methods from the literature. The results of the experiments show that the proposed filter-importance criterion outperforms the standard L1 and L2 approaches from the literature. Furthermore, the results also suggest that: 1) the pruned models are able to retain (or even improve on) the performance of the unpruned originals, as long as they are not over-pruned, with RITnet and U-Net at 50% of their original FLOPs reaching up to 4% and 7% higher IoU values than their unpruned versions, respectively, 2) smaller models require more careful pruning, as the pruning process can hurt the model’s generalization capabilities, and 3) the novel criterion most convincingly outperforms the classic approaches when sufficient training data is available, implying that the abundance of data leads to more robust activation-based importance computation.},

keywords = {biometrics, CNN, deep learning, model compression, pruning, sclera, sclera segmentation},

pubstate = {published},

tppubtype = {article}

}

@article{AccessMartin2023,

title = {ChildNet: Structural Kinship Face Synthesis Model With Appearance Control Mechanisms},

author = {Martin Pernuš and Mansi Bhatnagar and Badr Samad and Divyanshu Singh and Peter Peer and Vitomir Štruc and Simon Dobrišek},

url = {https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=10126110},

doi = {10.1109/ACCESS.2023.3276877},

issn = {2169-3536},

year  = {2023},

date = {2023-05-17},

journal = {IEEE Access},

pages = {1-22},

abstract = {Kinship face synthesis is an increasingly popular topic within the computer vision community, particularly the task of predicting the child appearance using parental images. Previous work has been limited in terms of model capacity and inadequate training data, which is comprised of low-resolution and tightly cropped images, leading to lower synthesis quality. In this paper, we propose ChildNet,  a method for kinship face synthesis that leverages the facial image generation capabilities of a state-of-the-art Generative Adversarial Network (GAN), and resolves the aforementioned problems. ChildNet is designed within the GAN latent space and is able to predict a child appearance that bears high resemblance to real parents’ children. To ensure fine-grained control, we propose an age and gender manipulation module that allows precise manipulation of the child synthesis result. ChildNet is capable of generating multiple child images per parent pair input, while providing a way to control the image generation variability. Additionally, we introduce a mechanism to control the dominant parent image. Finally, to facilitate the task of kinship face synthesis, we introduce a new kinship dataset, called Next of Kin. This dataset contains 3690 high-resolution face images with a diverse range of ethnicities and ages. We evaluate ChildNet in comprehensive experiments against three competing kinship face synthesis models, using two kinship datasets. The experiments demonstrate the superior performance of ChildNet in terms of identity similarity, while exhibiting high perceptual image quality. The source code for the model is publicly available at: https://github.com/MartinPernus/ChildNet.},

keywords = {artificial intelligence, CNN, deep learning, face generation, face synthesis, GAN, GAN inversion, kinship, kinship synthesis, StyleGAN2},

pubstate = {published},

tppubtype = {article}

}

@article{FadiIVCSynthetic,

title = {Synthetic data for face recognition: Current state and future prospects},

author = {Fadi Boutros and Vitomir Štruc and Julian Fierrez and Naser Damer},

url = {https://www.sciencedirect.com/science/article/pii/S0262885623000628},

doi = {https://doi.org/10.1016/j.imavis.2023.104688},

year  = {2023},

date = {2023-05-15},

urldate = {2023-05-15},

journal = {Image and Vision Computing},

number = {104688},

abstract = {Over the past years, deep learning capabilities and the availability of large-scale training datasets advanced rapidly, leading to breakthroughs in face recognition accuracy. However, these technologies are foreseen to face a major challenge in the next years due to the legal and ethical concerns about using authentic biometric data in AI model training and evaluation along with increasingly utilizing data-hungry state-of-the-art deep learning models. With the recent advances in deep generative models and their success in generating realistic and high-resolution synthetic image data, privacy-friendly synthetic data has been recently proposed as an alternative to privacy-sensitive authentic data to overcome the challenges of using authentic data in face recognition development. This work aims at providing a clear and structured picture of the use-cases taxonomy of synthetic face data in face recognition along with the recent emerging advances of face recognition models developed on the bases of synthetic data. We also discuss the challenges facing the use of synthetic data in face recognition development and several future prospects of synthetic data in the domain of face recognition.},

keywords = {biometrics, CNN, diffusion, face recognition, generative models, survey, synthetic data},

pubstate = {published},

tppubtype = {article}

}

@article{MedenDeID2023,

title = {Face deidentification with controllable privacy protection},

author = {Blaž Meden and Manfred Gonzalez-Hernandez and Peter Peer and Vitomir Štruc},

url = {https://reader.elsevier.com/reader/sd/pii/S0262885623000525?token=BC1E21411C50118E666720B002A89C9EB3DB4CFEEB5EB18D7BD7B0613085030A96621C8364583BFE7BAE025BE3646096&originRegion=eu-west-1&originCreation=20230516115322},

doi = {https://doi.org/10.1016/j.imavis.2023.104678},

year  = {2023},

date = {2023-04-01},

journal = {Image and Vision Computing},

volume = {134},

number = {104678},

pages = {1-19},

abstract = {Privacy protection has become a crucial concern in today’s digital age. Particularly sensitive here are facial images, which typically not only reveal a person’s identity, but also other sensitive personal information. To address this problem, various face deidentification techniques have been presented in the literature. These techniques try to remove or obscure personal information from facial images while still preserving their usefulness for further analysis. While a considerable amount of work has been proposed on face deidentification, most state-of-theart solutions still suffer from various drawbacks, and (a) deidentify only a narrow facial area, leaving potentially important contextual information unprotected, (b) modify facial images to such degrees, that image naturalness and facial diversity is suffering in the deidentify images, (c) offer no flexibility in the level of privacy protection ensured, leading to suboptimal deployment in various applications, and (d) often offer an unsatisfactory tradeoff between the ability to obscure identity information, quality and naturalness of the deidentified images, and sufficient utility preservation. In this paper, we address these shortcomings with a novel controllable face deidentification technique that balances image quality, identity protection, and data utility for further analysis. The proposed approach utilizes a powerful generative model (StyleGAN2), multiple auxiliary classification models, and carefully designed constraints to guide the deidentification process. The approach is validated across four diverse datasets (CelebA-HQ, RaFD, XM2VTS, AffectNet) and in comparison to 7 state-of-the-art competitors. The results of the experiments demonstrate that the proposed solution leads to: (a) a considerable level of identity protection, (b) valuable preservation of data utility, (c) sufficient diversity among the deidentified faces, and (d) encouraging overall performance.},

keywords = {CNN, deep learning, deidentification, face recognition, GAN, GAN inversion, privacy, privacy protection, StyleGAN2},

pubstate = {published},

tppubtype = {article}

}

@article{UhljiIETZiga,

title = {Efficient ear alignment using a two-stack hourglass network},

author = {Anja Hrovatič and Peter Peer and Vitomir Štruc and Žiga Emeršič},

url = {https://ietresearch.onlinelibrary.wiley.com/doi/epdf/10.1049/bme2.12109},

doi = {10.1049/bme2.12109},

issn = {2047-4938},

year  = {2023},

date = {2023-01-01},

journal = {IET Biometrics },

pages = {1-14},

abstract = {Ear images have been shown to be a reliable modality for biometric recognition with desirable characteristics, such as high universality, distinctiveness, measurability and permanence. While a considerable amount of research has been directed towards ear recognition techniques, the problem of ear alignment is still under-explored in the open literature. Nonetheless, accurate alignment of ear images, especially in unconstrained acquisition scenarios, where the ear appearance is expected to vary widely due to pose and view point variations, is critical for the performance of all downstream tasks, including ear recognition. Here, the authors address this problem and present a framework for ear alignment that relies on a two-step procedure: (i) automatic landmark detection and (ii) fiducial point alignment. For the first (landmark detection) step, the authors implement and train a Two-Stack Hourglass model (2-SHGNet) capable of accurately predicting 55 landmarks on diverse ear images captured in uncontrolled conditions. For the second (alignment) step, the authors use the Random Sample Consensus (RANSAC) algorithm to align the estimated landmark/fiducial points with a pre-defined ear shape (i.e. a collection of average ear landmark positions). The authors evaluate the proposed framework in comprehensive experiments on the AWEx and ITWE datasets and show that the 2-SHGNet model leads to more accurate landmark predictions than competing state-of-the-art models from the literature. Furthermore, the authors also demonstrate that the alignment step significantly improves recognition accuracy with ear images from unconstrained environments compared to unaligned imagery.},

keywords = {biometrics, CNN, deep learning, ear, ear alignment, ear recognition},

pubstate = {published},

tppubtype = {article}

}

@article{TextEmotionESWA,

title = {DHF-Net: A hierarchical feature interactive fusion network for dialogue emotion recognition},

author = {Chenquan Gan and Yucheng Yang and Qingyi Zhub and Deepak Kumar Jain and Vitomir Struc},

url = {https://www.sciencedirect.com/science/article/pii/S0957417422016025?via%3Dihub},

doi = {https://doi.org/10.1016/j.eswa.2022.118525},

year  = {2022},

date = {2022-12-30},

urldate = {2022-08-01},

journal = {Expert Systems with Applications},

volume = {210},

abstract = {To balance the trade-off between contextual information and fine-grained information in identifying specific emotions during a dialogue and combine the interaction of hierarchical feature related information, this paper proposes a hierarchical feature interactive fusion network (named DHF-Net), which not only can retain the integrity of the context sequence information but also can extract more fine-grained information. To obtain a deep semantic information, DHF-Net processes the task of recognizing dialogue emotion and dialogue act/intent separately, and then learns the cross-impact of two tasks through collaborative attention. Also, a bidirectional gate recurrent unit (Bi-GRU) connected hybrid convolutional neural network (CNN) group method is designed, by which the sequence information is smoothly sent to the multi-level local information layers for feature exaction. Experimental results show that, on two open session datasets, the performance of DHF-Net is improved by 1.8% and 1.2%, respectively.},

keywords = {attention, CNN, deep learning, dialogue, emotion recognition, fusion, fusion network, nlp, semantics, text, text processing},

pubstate = {published},

tppubtype = {article}

}

@article{TomasevicSensors,

title = {Reconstructing Superquadrics from Intensity and Color Images},

author = {Darian Tomašecić and Peter Peer and Franc Solina and Aleš Jaklič and Vitomir Štruc},

url = {https://www.mdpi.com/1424-8220/22/14/5332/pdf?version=1658380987},

doi = {https://doi.org/10.3390/s22145332},

year  = {2022},

date = {2022-07-16},

journal = {Sensors},

volume = {22},

number = {5332},

issue = {4},

abstract = {The task of reconstructing 3D scenes based on visual data represents a longstanding problem in computer vision. Common reconstruction approaches rely on the use of multiple volumetric primitives to describe complex objects. Superquadrics (a class of volumetric primitives) have shown great promise due to their ability to describe various shapes with only a few parameters. Recent research has shown that deep learning methods can be used to accurately reconstruct random superquadrics from both 3D point cloud data and simple depth images. In this paper, we extended these reconstruction methods to intensity and color images. Specifically, we used a dedicated convolutional neural network (CNN) model to reconstruct a single superquadric from the given input image. We analyzed the results in a qualitative and quantitative manner, by visualizing reconstructed superquadrics as well as observing error and accuracy distributions of predictions. We showed that a CNN model designed around a simple ResNet backbone can be used to accurately reconstruct superquadrics from images containing one object, but only if one of the spatial parameters is fixed or if it can be determined from other image characteristics, e.g., shadows. Furthermore, we experimented with images of increasing complexity, for example, by adding textures, and observed that the results degraded only slightly. In addition, we show that our model outperforms the current state-of-the-art method on the studied task. Our final result is a highly accurate superquadric reconstruction model, which can also reconstruct superquadrics from real images of simple objects, without additional training.},

keywords = {arrs, CNN, depth data, depth estimation, depth sensing, intensity images, superquadric, superquadrics},

pubstate = {published},

tppubtype = {article}

}

@article{CrowdCountingPevec,

title = {Benchmarking Crowd-Counting Techniques across Image Characteristics},

author = {Klemen Pevec and Klemen Grm and Vitomir Štruc},

url = {https://lmi.fe.uni-lj.si/wp-content/uploads/2022/10/Pevec.pdf

https://ev.fe.uni-lj.si/5-2021/Pevec.pdf},

year  = {2021},

date = {2021-05-01},

journal = {Elektorethniski Vestnik},

volume = {88},

issue = {5},

pages = {227-235},

abstract = {Crowd--counting is a longstanding computer vision used in estimating the crowd sizes for security purposes at public protests in streets, public gatherings, for collecting crowd statistics at airports, malls, concerts, conferences, and other similar venues, and for monitoring people and crowds during public health crises (such as the one caused by COVID-19). Recently, the performance of automated methods for crowd--counting from single images has improved particularly due to the introduction of deep learning techniques and large labelled training datasets. However, the robustness of these methods to varying imaging conditions, such as weather, image perspective, and large variations in the crowd size has not been studied in-depth in the open literature. To address this gap, a systematic study on the robustness of four recently developed crowd--counting methods is performed in this paper to evaluate their performance with respect to variable (real-life) imaging scenarios that include different event types, weather conditions, image sources and crowd sizes. It is shown that the performance of the tested techniques is degraded in unclear weather conditions (i.e., fog, rain, snow) and also on images taken from large distances by drones. On the opposite, clear weather conditions, crowd--counting methods can provide accurate and usable results.},

keywords = {CNN, crowd counting, drones, image characteristics, model comparison, neural networks},

pubstate = {published},

tppubtype = {article}

}

@article{TIPKlemen_2020,

title = {Face hallucination using cascaded super-resolution and identity priors},

author = {Klemen Grm and Walter J. Scheirer and Vitomir Štruc},

url = {https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8866753

https://lmi.fe.uni-lj.si/wp-content/uploads/2023/02/IEEET_face_hallucination_compressed.pdf},

doi = {10.1109/TIP.2019.2945835},

year  = {2020},

date = {2020-01-01},

urldate = {2020-01-01},

journal = {IEEE Transactions on Image Processing},

abstract = {In this paper we address the problem of hallucinating high-resolution facial images from low-resolution inputs at high magnification factors. We approach this task with convolutional neural networks (CNNs) and propose a novel (deep) face hallucination model that incorporates identity priors into the learning procedure. The model consists of two main parts: i) a cascaded super-resolution network that upscales the lowresolution facial images, and ii) an ensemble of face recognition models that act as identity priors for the super-resolution network during training. Different from most competing super-resolution techniques that rely on a single model for upscaling (even with large magnification factors), our network uses a cascade of multiple SR models that progressively upscale the low-resolution images using steps of 2×. This characteristic allows us to apply supervision signals (target appearances) at different resolutions and incorporate identity constraints at multiple-scales. The proposed C-SRIP model (Cascaded Super Resolution with Identity Priors) is able to upscale (tiny) low-resolution images captured in unconstrained conditions and produce visually convincing results for diverse low-resolution inputs. We rigorously evaluate the proposed model on the Labeled Faces in the Wild (LFW), Helen and CelebA datasets and report superior performance compared to the existing state-of-the-art.

},

keywords = {biometrics, CNN, computer vision, deep learning, face, face hallucination, super-resolution},

pubstate = {published},

tppubtype = {article}

}

@article{vitek2020comprehensive,

title = {A comprehensive investigation into sclera biometrics: a novel dataset and performance study},

author = {Matej Vitek and Peter Rot and Vitomir Struc and Peter Peer},

url = {https://link.springer.com/epdf/10.1007/s00521-020-04782-1},

doi = {https://doi.org/10.1007/s00521-020-04782-1},

year  = {2020},

date = {2020-01-01},

journal = {Neural Computing and Applications},

pages = {1-15},

abstract = {The area of ocular biometrics is among the most popular branches of biometric recognition technology. This area has long been dominated by iris recognition research, while other ocular modalities such as the periocular region or the vasculature of the sclera have received significantly less attention in the literature. Consequently, ocular modalities beyond the iris are not well studied and their characteristics are today still not as well understood. While recent needs for more secure authentication schemes have considerably increased the interest in competing ocular modalities, progress in these areas is still held back by the lack of publicly available datasets that would allow for more targeted research into specific ocular characteristics next to the iris. In this paper, we aim to bridge this gap for the case of sclera biometrics and introduce a novel dataset designed for research into ocular biometrics and most importantly for research into the vasculature of the sclera. Our dataset, called Sclera Blood Vessels, Periocular and Iris (SBVPI), is, to the best of our knowledge, the first publicly available dataset designed specifically with research in sclera biometrics in mind. The dataset contains high-quality RGB ocular images, captured in the visible spectrum, belonging to 55 subjects. Unlike competing datasets, it comes with manual markups of various eye regions, such as the iris, pupil, canthus or eyelashes and a detailed pixel-wise annotation of the complete sclera vasculature for a subset of the images. Additionally, the datasets ship with gender and age labels. The unique characteristics of the dataset allow us to study aspects of sclera biometrics technology that have not been studied before in the literature (e.g. vasculature segmentation techniques) as well as issues that are of key importance for practical recognition systems. Thus, next to the SBVPI dataset we also present in this paper a comprehensive investigation into sclera biometrics and the main covariates that affect the performance of sclera segmentation and recognition techniques, such as gender, age, gaze direction or image resolution. Our experiments not only demonstrate the usefulness of the newly introduced dataset, but also contribute to a better understanding of sclera biometrics in general.},

keywords = {biometrics, CNN, dataset, multi-view, ocular, performance study, recognition, sclera, segmentation, visible light},

pubstate = {published},

tppubtype = {article}

}

@article{grm2017strengths,

title = {Strengths and weaknesses of deep learning models for face recognition against image degradations},

author = {Klemen Grm and Vitomir Štruc and Anais Artiges and Matthieu Caron and Hazim K. Ekenel},

url = {https://arxiv.org/pdf/1710.01494.pdf},

year  = {2017},

date = {2017-01-01},

journal = {IET Biometrics},

volume = {7},

number = {1},

pages = {81--89},

publisher = {IET},

abstract = {Convolutional neural network (CNN) based approaches are the state of the 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 the strengths and weaknesses of existing deep models for face recognition are still relatively scarce. In this paper, we try to fill this gap and study the effects of different covariates on the verification performance of four recent CNN models using the Labelled Faces in the Wild dataset. Specifically, we investigate the influence of covariates related to image quality and model characteristics, and analyse their impact on the face verification performance of different deep CNN models. Based on comprehensive and rigorous experimentation, we identify the strengths and weaknesses of the deep learning models, and present key areas for potential future research. Our results indicate that high levels of noise, blur, missing pixels, and brightness have a detrimental effect on the verification performance of all models, whereas the impact of contrast changes and compression artefacts is limited. We find that the descriptor-computation strategy and colour information does not have a significant influence on performance.},

keywords = {CNN, convolutional neural networks, face recognition, googlenet, study, vgg},

pubstate = {published},

tppubtype = {article}

}