Koşmaz Sünnetci, Betül; Bingöl, Özkan; Gedikli, Eyüp; Ekinci, Murat; Doğan, Ramazan Özgür; Türk, Salih; Güngör, Nihan

Featured Application: The proposed framework facilitates secure and user-friendly biometric authentication on mobile platforms, particularly in mobile payment and access control scenarios, under unconstrained conditions. The widespread adoption of mobile devices has made secure and user-friendly biometric authentication critical. However, widely used modalities such as fingerprint and facial recognition show limited robustness under uncontrolled illumination and on heterogeneous devices. In contrast, palmprint recognition offers strong potential because of its rich textural patterns and high discriminative power. This study addresses the limitations of laboratory-based datasets that fail to capture real-world challenges. We introduce MPW-180, a novel dataset comprising videos of 180 participants recorded on their own smartphones in everyday environments. By systematically incorporating diverse illumination conditions (with and without flash) and natural free-hand movements, MPW-180 is the first dataset to adopt a bring-your-own-device paradigm, providing a realistic benchmark for evaluating generalization in mobile biometric models. In addition, we propose PalmWildNet, an SE-block-enhanced deep learning architecture trained with Triplet Loss and a cross-illumination sampling strategy. The experimental results show that conventional methods suffer over 50% performance degradation under cross-illumination conditions. In contrast, our method reduces the Equal Error Rate to 1–2% while maintaining an accuracy above 97%. These findings demonstrate that the proposed framework not only tolerates illumination variability but also learns robust illumination-invariant representations, making it well-suited for mobile biometric authentication.

Sünnetci Koşmaz, Betül; Bingöl, Özkan; Gedikli, Eyüp; Ekinci, Murat

In palm biometric recognition systems, the hand being at different angles and positions relative to the camera presents a critical challenge in terms of accuracy and reliability. Especially in images captured in free environments, hand misalignment can negatively affect recognition performance by making it difficult to extract biometric attributes. Perspective distortions due to camera position change the geometric structure of the hand, compromising biometric data integrity. While 3D imaging systems can offer successful solutions for pose normalization, performing a similar transformation for 2D images poses a major technical challenge. In this study, depth estimation supported plane fitting is used to remove perspective distortions in 2D images and pose normalization of the hand is obtained. The proposed method uses the information obtained from depth estimation to realign the hand image as if it were standing directly in front of the camera, improving geometric consistency. This will make the extraction of biometric features more reliable and contribute to improving the accuracy of palm recognition systems.

Yaprak, Büşranur; Gedikli, Eyüp

Gait, a behavior-based biometric feature, has gained increasing popularity in human identification, particularly in surveillance systems, due to its ability to function without physical contact or explicit consent. Traditional silhouette-based methods have demonstrated that different body parts exhibit distinct movement patterns during walking, thereby enhancing recognition accuracy. In this study, we propose an improved part-based gait recognition approach by leveraging ensemble learning on local body regions. The Gait Energy Image (GEI) is segmented into five horizontal parts, and ensemble learning is applied to the convolutional neural network (CNN) responsible for their processing. A separate MetaModel is trained for each body part to integrate the part-based features obtained from ensemble learning and synthesize the most discriminative ones. Additionally, a part-removal process is introduced to mitigate the effects of appearance-based variations by analyzing absolute differences between images with and without variations. The aggregated most distinctive features contribute to robust recognition. We evaluate our proposed approach on the CASIA-B, CASIA-C, and Outdoor-Gait datasets, and experimental results indicate that ensemble learning significantly enhances part-based gait recognition performance under various appearance variations, outperforming several state-of-the-art methods. The datasets and source code are available at https://github.com/busrakckugurlu/Enhancing-Part-based-Gait-Recognition-via-Ensemble-Learning-and-Feature-Fusion/tree/main.

Yaprak, Büşranur; Gedikli, Eyüp

Gait recognition is the process of identifying a person from a distance based on their walking patterns. However, the recognition rate drops significantly under cross-view angle and appearance-based variations. In this study, the effectiveness of the most well-known gait representations in solving this problem is investigated based on deep learning. For this purpose, a comprehensive performance evaluation is performed by combining different modalities, including silhouettes, optical flows, and concatenated image of the Gait Energy Image (GEI) head and leg region, with GEI itself. This evaluation is carried out across different multimodal deep convolutional neural network (CNN) architectures, namely fine-tuned EfficientNet-B0, MobileNet-V1, and ConvNeXt-base models. These models are trained separately on GEIs, silhouettes, optical flows, and concatenated image of GEI head and leg regions, and then extracted GEI features are fused in pairs with other extracted modality features to find the most effective gait combination. Experimental results on the two different datasets CASIA-B and Outdoor-Gait show that the concatenated image of GEI head and leg regions significantly increased the recognition rate of the networks compared to other modalities. Moreover, this modality demonstrates greater robustness under varied carrying (BG) and clothing (CL) conditions compared to optical flows (OF) and silhouettes (SF). Codes available at https://github.com/busrakckugurlu/Different-gait-combinations-based-on-multi-modal-deep-CNN-architectures.git

Küçükuğurlu, Büşranur; Gedikli, Eyüp

Thresholding is a frequently used method in image processing because of its consistency and low computational cost. Otsu's and Kapur's methods are two important techniques that were proved to be best thresholding methods. However, they have high computational complexity when extended to multilevel thresholding because of their exhaustively search. Recently, meta-heuristic algorithms have been successfully applied for thresholding problems. In this study, six different meta-heuristic algorithms based on Otsu's and Kapur's functions; Particle Swarm Optimization (PSO), Firefly Algorithm (FA), Symbiotic Organisms Search (SOS), Artifical Bee Colony (ABC), Genetic Algorithm (GA) and grey Wolf Optimizer (GWO) were used for multilevel thresholding problem and compared. Experimental results suggest that SOS, PSO and FA algorithms often have higher fitness values than other algorithms. Especially when more than two threshold values are determined, SOS algorithm mostly gives higher fitness values. PSNR and SSIM results of the algorithms are similar. In terms of computational complexity, the GWO algorithm has the fastest convergence. For standard deviations of objective functions; more stable results were obtained with SOS based on Kapur's function, SOS and PSO based on Otsu's function. Also, SOS based on Kapur's function was found to be the most successful algorithm in the Friedman test. As a result, although the GWO approached faster, the SOS algorithm produced more consistent results for both objective functions.