Ahmed Taha
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ahmed Taha has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
El-Sayed M. El-Horbaty
3 shared publicationsHala H. Zayed
3 shared publications- Hala H. Zayed
Computer Science Department, Faculty of Computers and Artificial Intelligence, Benha University, Benha, Egypt
3 shared publications M. E. Khalifa
3 shared publications- Eman I. Abd El-Latif
Faculty of Science, Benha University, Benha, Egypt
2 shared publications - Mai K. Galab
Department of Computer Science, Faculty of Computers and Artificial Intelligence, Benha University, Benha 13518, Egypt
2 shared publications
Latest Publications
- Open Access
Automatic Non-Invasive Jaundice Detection of Neonates Using Computer Vision
Smartphone-based screening for neonatal jaundice offers a low-cost alternative to conventional bilirubin measurement. Still, most computer vision techniques rely on color calibration, fixed skin segmentation, or a limited set of features, which reduces their reliability across different imaging conditions. In this work, we propose a calibration-free, smartphone-based approach for assessing neonatal jaundice by combining adaptive skin segmentation, multi-color feature extraction, SHAP-guided feature selection, and machine learning. Our framework uses visual features from the images to direct the process to either SAM or DensePose to extract the skin from public images or more complex clinical images. The skin areas are enhanced and represented in the RGB, HSV, CIE Lab, and YCbCr color spaces, and the chromatic features are then ranked and selected using SHAP. The selected features were evaluated using machine learning algorithms to estimate TsB continuously and to perform three-class jaundice classification. The experiments were carried out on the available NeoJaundice dataset and a clinical NICU dataset collected at Benha University Hospital. The segmentation evaluation supported adaptive segmentation, with Dice/IoU scores of 0.8807/0.7903 for DensePose (Benha) and 0.9850/0.9705 for SAM (NeoJaundice). For TSB estimation, KNN obtained an R of 0.982, an R2 of 0.965, and an RMSE of 0.973 mg/dL. For the three-class classification, Random Forest achieved accuracies of 98.31% on the NeoJaundice dataset and 98.44% on the clinical dataset. When evaluated on 261 clinical images not previously seen, an accuracy of 98.47% was achieved, providing evidence of its performance on previously unseen NICU images. The results show the promise of integrating adaptive segmentation, compact multi-color representations, and explainable feature selection for calibration-free smartphone-based screening of neonatal jaundice. Further multicenter validation across diverse neonatal populations, skin tones, and smartphone devices is needed to establish broader clinical applicability.
Open AccessAutomatic Non-Invasive Jaundice Detection of Neonates Using Computer Vision
Smartphone-based screening for neonatal jaundice offers a low-cost alternative to conventional bilirubin measurement. Still, most computer vision techniques rely on color calibration, fixed skin segmentation, or a limited set of features, which reduces their reliability across different imaging conditions. In this work, we propose a calibration-free, smartphone-based approach for assessing neonatal jaundice by combining adaptive skin segmentation, multi-color feature extraction, SHAP-guided feature selection, and machine learning. Our framework uses visual features from the images to direct the process to either SAM or DensePose to extract the skin from public images or more complex clinical images. The skin areas are enhanced and represented in the RGB, HSV, CIE Lab, and YCbCr color spaces, and the chromatic features are then ranked and selected using SHAP. The selected features were evaluated using machine learning algorithms to estimate TsB continuously and to perform three-class jaundice classification. The experiments were carried out on the available NeoJaundice dataset and a clinical NICU dataset collected at Benha University Hospital. The segmentation evaluation supported adaptive segmentation, with Dice/IoU scores of 0.8807/0.7903 for DensePose (Benha) and 0.9850/0.9705 for SAM (NeoJaundice). For TSB estimation, KNN obtained an R of 0.982, an R2 of 0.965, and an RMSE of 0.973 mg/dL. For the three-class classification, Random Forest achieved accuracies of 98.31% on the NeoJaundice dataset and 98.44% on the clinical dataset. When evaluated on 261 clinical images not previously seen, an accuracy of 98.47% was achieved, providing evidence of its performance on previously unseen NICU images. The results show the promise of integrating adaptive segmentation, compact multi-color representations, and explainable feature selection for calibration-free smartphone-based screening of neonatal jaundice. Further multicenter validation across diverse neonatal populations, skin tones, and smartphone devices is needed to establish broader clinical applicability.
- Open Access
- 5 August 2026
- Published by Springer Nature in Journal of Big Data
- Vol. 13 (1) , 139
An optimized approach for brain tumor classification using MobileNet and enhanced Newton-Raphson optimizer
This article introduces an enhanced version of the Newton-Raphson-based optimizer (I-NRBO) with a complex optimization technique termed the Local Escaping Operator (LEO). This method effectively balances solution variety with convergence speed, enhancing optimization effectiveness and reducing the likelihood of local optima. The research presents an advanced and efficient model that utilizes a pre-trained deep learning architecture, MobileNet, with the I-NRBO algorithm to enhance the classification accuracy of brain tumors. The I-NRBO method is evaluated over CEC-2022, the results showed that I-NRBO outperformed the traditional NRBO and other leading algorithms in terms of statistical convergence and various criteria. In the subsequent study, the proposed I-NRBO is applied to optimize hyperparameters in the MobileNet model, specifically for brain tumor classification. The I-NRBO-MobileNet model is assessed against many previous research and alternative deep learning architectures with comparable parameters, including VGG16, ResNet50, InceptionV3, and MobileNet as baseline models. The experimental findings reveal that the I-NRBO-MobileNet model outstanding performance, with an accuracy of 99.24%, a precision of 99.25%, a recall of 99.24%, an AUC of 99.95%, and a specificity of 99.70%. Results demonstrate that the suggested I-NRBO-MobileNet model can accurately classify brain cancers.
- 5 August 2026
- Published by Springer Nature in Journal of Big Data
- Vol. 13 (1) , 139
Open AccessAn optimized approach for brain tumor classification using MobileNet and enhanced Newton-Raphson optimizer
This article introduces an enhanced version of the Newton-Raphson-based optimizer (I-NRBO) with a complex optimization technique termed the Local Escaping Operator (LEO). This method effectively balances solution variety with convergence speed, enhancing optimization effectiveness and reducing the likelihood of local optima. The research presents an advanced and efficient model that utilizes a pre-trained deep learning architecture, MobileNet, with the I-NRBO algorithm to enhance the classification accuracy of brain tumors. The I-NRBO method is evaluated over CEC-2022, the results showed that I-NRBO outperformed the traditional NRBO and other leading algorithms in terms of statistical convergence and various criteria. In the subsequent study, the proposed I-NRBO is applied to optimize hyperparameters in the MobileNet model, specifically for brain tumor classification. The I-NRBO-MobileNet model is assessed against many previous research and alternative deep learning architectures with comparable parameters, including VGG16, ResNet50, InceptionV3, and MobileNet as baseline models. The experimental findings reveal that the I-NRBO-MobileNet model outstanding performance, with an accuracy of 99.24%, a precision of 99.25%, a recall of 99.24%, an AUC of 99.95%, and a specificity of 99.70%. Results demonstrate that the suggested I-NRBO-MobileNet model can accurately classify brain cancers.
- Open Access
- 15 December 2025
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 51 (9) , 12147-12166
DeepRetina Framework for Multi-retinal Diseases Classification
Diagnosing retinal diseases is a fundamental challenge in developing robust multi-disease classification systems due to the limited availability of datasets and inconsistent quality. Therefore, this study presents $$\text{DeepRetina}$$ , a framework addressing these challenges through dataset harmonization. Five distinct retinal image datasets were unified using systematic preprocessing, resulting in a consolidated dataset of 29,966 high-resolution fundus images across eight disease categories. Harmonization corrects variations in image quality, color, and lighting resulting from different imaging devices or conditions. Data harmonization enhances the model’s ability to generalize across diverse datasets by standardizing the color and texture properties of images. The study compares the performance of custom CNN, EfficientNetV2, and MobileNetV3Large architectures for multi-disease classification. EfficientNetV2 achieved the highest accuracy of 79% with a precision of 54%. The proposed methodology significantly advances the field by (1) establishing a robust approach for harmonizing heterogeneous datasets, (2) presenting a large-scale, unified dataset for future research, and (3) presenting a comparative analysis of deep learning architectures optimized for retinal disease classification. $$\text{DeepRetina}$$ lays the foundation for scalable and accurate automated retinal disease diagnosis, contributing to improved detection and classification in ophthalmology.- 15 December 2025
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 51 (9) , 12147-12166
Open AccessDeepRetina Framework for Multi-retinal Diseases Classification
Diagnosing retinal diseases is a fundamental challenge in developing robust multi-disease classification systems due to the limited availability of datasets and inconsistent quality. Therefore, this study presents $$\text{DeepRetina}$$ , a framework addressing these challenges through dataset harmonization. Five distinct retinal image datasets were unified using systematic preprocessing, resulting in a consolidated dataset of 29,966 high-resolution fundus images across eight disease categories. Harmonization corrects variations in image quality, color, and lighting resulting from different imaging devices or conditions. Data harmonization enhances the model’s ability to generalize across diverse datasets by standardizing the color and texture properties of images. The study compares the performance of custom CNN, EfficientNetV2, and MobileNetV3Large architectures for multi-disease classification. EfficientNetV2 achieved the highest accuracy of 79% with a precision of 54%. The proposed methodology significantly advances the field by (1) establishing a robust approach for harmonizing heterogeneous datasets, (2) presenting a large-scale, unified dataset for future research, and (3) presenting a comparative analysis of deep learning architectures optimized for retinal disease classification. $$\text{DeepRetina}$$ lays the foundation for scalable and accurate automated retinal disease diagnosis, contributing to improved detection and classification in ophthalmology. - Open Access
- 1 December 2025
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 14 (6)
Improved copy-move forgery detection through multilevel clustering
Copy move forgery detection (CMFD) based on keypoints remains a widely used technique; however, it often struggles to effectively identify small and smoothly tampered regions within images. This paper introduces a CMFD method that enhances detection accuracy by integrating a double-matching process with advanced region localization techniques. Delaunay triangles formed by accelerated KAZE (AKAZE) and scale-invariant feature transform (SIFT) features are matched in the double-matching process to identify suspicious regions. To ensure sufficient keypoint pairs, the set of matching triangles is iteratively expanded to include neighboring triangles, covering the entire tampered area. Subsequently, a second matching with a looser threshold is performed on the vertices. In the region localization process, the multilevel density-based spatial clustering of applications with noise (DBSCAN) effectively handles scenarios involving multiple copied regions with varying sizes. Using the standard MICC-F600 and COVERAGE datasets, experiments demonstrate that the proposed CMFD method is robust and achieves better performance than state-of-the-art baselines.
- 1 December 2025
- Published by Institute of Advanced Engineering and Science in IAES International Journal of Artificial Intelligence (IJ-AI)
- Vol. 14 (6)
Open AccessImproved copy-move forgery detection through multilevel clustering
Copy move forgery detection (CMFD) based on keypoints remains a widely used technique; however, it often struggles to effectively identify small and smoothly tampered regions within images. This paper introduces a CMFD method that enhances detection accuracy by integrating a double-matching process with advanced region localization techniques. Delaunay triangles formed by accelerated KAZE (AKAZE) and scale-invariant feature transform (SIFT) features are matched in the double-matching process to identify suspicious regions. To ensure sufficient keypoint pairs, the set of matching triangles is iteratively expanded to include neighboring triangles, covering the entire tampered area. Subsequently, a second matching with a looser threshold is performed on the vertices. In the region localization process, the multilevel density-based spatial clustering of applications with noise (DBSCAN) effectively handles scenarios involving multiple copied regions with varying sizes. Using the standard MICC-F600 and COVERAGE datasets, experiments demonstrate that the proposed CMFD method is robust and achieves better performance than state-of-the-art baselines.
- 22 August 2025
- Published by Egyptian Knowledge Bank in Advanced Sciences and Technology Journal
Smart Serve: Redefining Customer Support with AI-Driven Ticketing Intelligence
With growing customer expectations and increasing volumes of support requests, traditional helpdesk systems often fall short, resulting in slow resolution times, misrouted tickets, and a less-than-ideal customer experience. This paper presents Smart Serve, a cutting-edge AI-based ticketing solution that revolutionizes the working model of customer support teams. By leveraging cutting-edge Natural Language Processing (NLP) and state-of-the-art Large Language Models (LLMs), such as GPT and Gemini, the system automatically evaluates incoming requests for urgency, complexity, and category, enabling real-time prioritization, precise routing, and efficient handling. What differentiates this solution is its hybrid architecture: mundane issues are handled automatically with AI-powered responses, while tough ones are passed on smoothly to human agents. The combination of automation and human interaction will not only boost efficiency but also deliver a responsive and personalized customer experience. This study brings out the potential of innovative ticketing systems to revolutionize customer care, making it quicker, wiser, and more proactive.
- 22 August 2025
- Published by Egyptian Knowledge Bank in Advanced Sciences and Technology Journal
Smart Serve: Redefining Customer Support with AI-Driven Ticketing Intelligence
With growing customer expectations and increasing volumes of support requests, traditional helpdesk systems often fall short, resulting in slow resolution times, misrouted tickets, and a less-than-ideal customer experience. This paper presents Smart Serve, a cutting-edge AI-based ticketing solution that revolutionizes the working model of customer support teams. By leveraging cutting-edge Natural Language Processing (NLP) and state-of-the-art Large Language Models (LLMs), such as GPT and Gemini, the system automatically evaluates incoming requests for urgency, complexity, and category, enabling real-time prioritization, precise routing, and efficient handling. What differentiates this solution is its hybrid architecture: mundane issues are handled automatically with AI-powered responses, while tough ones are passed on smoothly to human agents. The combination of automation and human interaction will not only boost efficiency but also deliver a responsive and personalized customer experience. This study brings out the potential of innovative ticketing systems to revolutionize customer care, making it quicker, wiser, and more proactive.
- Open Access
- 1 October 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 36 (1) , 290-301
Tailoring AES for resource-constrained IoT devices
The internet of things (IoT) is a network of interconnected hardware, software, and many infrastructures that require cryptography solutions to provide security. IoT security is a critical concern, and it can be settled by using cryptographic algorithms such as advanced encryption standard (AES) for encryption and authentication. A fundamental component within the AES algorithm is the substitution box (S-box), which generates confusion and nonlinearity between plaintext and ciphertext, strengthening the process of security. This paper introduces a comparative analysis to offer valuable knowledge of the factors related to different S-box modifications, which will ultimately affect the design of cryptographic systems that use the AES algorithm. Then, a tailored AES algorithm is proposed for resource-constrained IoT devices by changing the standard S-box with another S-box. The new S-box reduces the rounds number and the time needed for the AES algorithm’s encryption, decryption, and key expansion. The performance of the proposed AES is assessed through various experiments. Therefore, our tailored AES with the new S-box is more secure and efficient than AES with a standard S-box.
- 1 October 2024
- Published by Institute of Advanced Engineering and Science in Indonesian Journal of Electrical Engineering and Computer Science
- Vol. 36 (1) , 290-301
Open AccessTailoring AES for resource-constrained IoT devices
The internet of things (IoT) is a network of interconnected hardware, software, and many infrastructures that require cryptography solutions to provide security. IoT security is a critical concern, and it can be settled by using cryptographic algorithms such as advanced encryption standard (AES) for encryption and authentication. A fundamental component within the AES algorithm is the substitution box (S-box), which generates confusion and nonlinearity between plaintext and ciphertext, strengthening the process of security. This paper introduces a comparative analysis to offer valuable knowledge of the factors related to different S-box modifications, which will ultimately affect the design of cryptographic systems that use the AES algorithm. Then, a tailored AES algorithm is proposed for resource-constrained IoT devices by changing the standard S-box with another S-box. The new S-box reduces the rounds number and the time needed for the AES algorithm’s encryption, decryption, and key expansion. The performance of the proposed AES is assessed through various experiments. Therefore, our tailored AES with the new S-box is more secure and efficient than AES with a standard S-box.
- 8 February 2021
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 46 (4) , 4049-4058
Adaptive Technique for Brightness Enhancement of Automated Knife Detection in Surveillance Video with Deep Learning
- 8 February 2021
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 46 (4) , 4049-4058
Adaptive Technique for Brightness Enhancement of Automated Knife Detection in Surveillance Video with Deep Learning
- 17 February 2020
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 45 (4) , 3379-3386
A Passive Approach for Detecting Image Splicing Based on Deep Learning and Wavelet Transform
- 17 February 2020
- Published by Springer Nature in Arabian Journal for Science and Engineering
- Vol. 45 (4) , 3379-3386
A Passive Approach for Detecting Image Splicing Based on Deep Learning and Wavelet Transform
- Open Access
- 8 May 2019
- Published by MECS Publisher in International Journal of Computer Network and Information Security
- Vol. 11 (5) , 28-35
A Passive Approach for Detecting Image Splicing using Deep Learning and Haar Wavelet Transform
Passive image forgery detection has attracted many researchers in the recent years. Image manipulation becomes easier than before because of the fast development of digital image editing software. Image splicing is one of the most widespread methods for tampering images. Research on detection of image splicing still carries great challenges. In this paper, an algorithm based on deep learning approach and wavelet transform is proposed to detect the spliced image. In the deep learning approach, Convolution Neural Network (CNN) is employed to automatically extract features from the spliced image. CNN is applied and then Haar Wavelet Transform (HWT) is used. Support Vector Machine (SVM) is used later for classification. Additional experiments are performed. That is, Discrete Cosine Transform (DCT) replaces HWT and then Principle Component Analysis (PCA) is applied. The proposed algorithm is evaluated on a publicly available image splicing datasets (CASIA v1.0 and CASIA v2.0). It achieves high accuracy while using a relatively low dimension feature vector. Our results demonstrate that the proposed algorithm is effective and accomplishes better performance for detecting the spliced image.
- 8 May 2019
- Published by MECS Publisher in International Journal of Computer Network and Information Security
- Vol. 11 (5) , 28-35
Open AccessA Passive Approach for Detecting Image Splicing using Deep Learning and Haar Wavelet Transform
Passive image forgery detection has attracted many researchers in the recent years. Image manipulation becomes easier than before because of the fast development of digital image editing software. Image splicing is one of the most widespread methods for tampering images. Research on detection of image splicing still carries great challenges. In this paper, an algorithm based on deep learning approach and wavelet transform is proposed to detect the spliced image. In the deep learning approach, Convolution Neural Network (CNN) is employed to automatically extract features from the spliced image. CNN is applied and then Haar Wavelet Transform (HWT) is used. Support Vector Machine (SVM) is used later for classification. Additional experiments are performed. That is, Discrete Cosine Transform (DCT) replaces HWT and then Principle Component Analysis (PCA) is applied. The proposed algorithm is evaluated on a publicly available image splicing datasets (CASIA v1.0 and CASIA v2.0). It achieves high accuracy while using a relatively low dimension feature vector. Our results demonstrate that the proposed algorithm is effective and accomplishes better performance for detecting the spliced image.
- 12 May 2015
- Published by Al-Zaytoonah University of Jordan
- p. 577-586
Human Activity Recognition for Surveillance Applications
- 12 May 2015
- Published by Al-Zaytoonah University of Jordan
- p. 577-586
Human Activity Recognition for Surveillance Applications