Muhammed Mutlu Yapici

Muhammed Mutlu Yapici

Auto-generated profile, based on publicly available information

Muhammed Mutlu Yapici

Computer Technologies Department, Ankara University, Ankara, Turkey

Research Focus & Metrics

Academic Career Length

9Years (2018-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
1 January 2022 in Computers & Security
Verified
Journal Article
Closed access
ATAdem Tekerek
Adem Tekerek
MYMuhammed Mutlu Yapici
Muhammed Mutlu Yapici

24 September 2020 in Pattern Analysis and Applications
Verified
Journal Article
Closed access
PDF file
MYMuhammed Mutlu Yapıcı
Muhammed Mutlu Yapıcı
ATAdem Tekerek
Adem Tekerek
NTNurettin Topaloğlu
Nurettin Topaloğlu

1 December 2018
Verified
Conference Paper
Closed access
MYMuhammed Mutlu Yapici
Muhammed Mutlu Yapici
ATAdem Tekerek
Adem Tekerek
NTNurettin Topaloglu
Nurettin Topaloglu
One of the most important biometric authentication technique is signature. Nowadays, there are two types of signatures, offline (static) and online (dynamic). Online signatures have higher distinctive features but offline signatures have fewer distinctive features. So offline signatures are more difficult to verify. In addition, the most important drawback of offline signatures is that they cannot be signed with the same way even by the most talented signer. This is called intra-personal variability. All these make the offline signature verification a challenging problem for researchers. In this study, we proposed a Deep Learning (DL) based offline signature verification method to prevent signature fraud by malicious people. The DL method used in the study is the Convolutional Neural Network (CNN). CNN was designed and trained separately for two different models such one Writer Dependent (WD) and the other Writer Independent (WI). The experimental results showed that WI has 62.5% of success and WD has 75% of success. It is predicted that the success of the obtained results will increase if the CNN method is supported by adding extra feature extraction methods.

26 May 2025 in Knowledge-Based Systems
Verified
Journal Article
Closed access
MYMuhammed Mutlu Yapici
Muhammed Mutlu Yapici

1 October 2019
Verified
Conference Paper
Closed access
MYMuhammed Mutlu Yapıcı
Muhammed Mutlu Yapıcı
ATAdem Tekerek
Adem Tekerek
NTNurettin Topaloğlu
Nurettin Topaloğlu
Deep learning methods are used in many popular areas such: image processing, computer vision, autonomous vehicles, character recognition, audio and video processing. These methods require high processing power, such as graphics cards (GPUs), to obtain successful results in the solution of NP hard problems which have big data. In this study, performance comparison of convolutional neural network (CNN) architectures were performed on GPU. ResNet, VGGNet19 and DenseNet CNN models, and GPDS signature dataset were used for comparison. According to the obtained results, ResNet50 took up the least amount of GPU memory space. The best classification results were obtained with DenseNet121 and the second one was from ResNet50.

26 February 2024 in IEEE Access
Verified
Journal Article
Open access
BKBurak Kucukgoz
Burak Kucukgoz
MYMuhammed Mutlu Yapici
Muhammed Mutlu Yapici
DMDeclan C. Murphy
Declan C. Murphy
ESEmma Spowart
Emma Spowart
DSDavid H Steel
David H Steel
BOBoguslaw Obara
Boguslaw Obara
This study presents a fully automated image informatics framework. The framework is combined with a deep learning (DL) approach to automatically predict visual acuity outcomes for people undergoing surgery for idiopathic full-thickness macular holes using 3D spectral-domain optical coherence tomography (SD-OCT) images. To overcome the impact of high variation in real-world image quality on the robustness of DL models, comprehensive imaging data pre-processing, quality assurance, and anomaly detection procedures were utilised. We then implemented, trained, and tested nine state-of-the-art DL predictive models through our designed loss function with multiple 2D input channels on the imaging dataset. Finally, we quantitatively compared the models using four evaluation metrics. Overall, the predictive model achieved a MAE of 6.47 ETDRS letters score, demonstrating high predictability. This confirms that our fully automated approach with input from seven central SD-OCT images from each patient can robustly predict visual acuity measurements. Further research will focus on adapting 3D DL-based predictive models and the uncertainty of 2D and 3D DL-based predictive models.