Jul 2026· International Journal for Research in Applied Science and Engineering Technology· Vol 14, pp. 2438-2446· 0 citations
TL;DR
This study builds and test a Deep Convolutional Generative Adversarial Network (DCGAN) that can produce realistic portraits of people's faces and proves that DCGANs are capable of creating realistic facial representations.
Abstract
Generative Adversarial Networks (GANs) are a potent tool for generating photorealistic synthetic images. In this study,
we aim to build and test a Deep Convolutional Generative Adversarial Network (DCGAN) that can produce realistic portraits of
people's faces. One network learns to transform random noise vectors into realistic face photographs, while the other learns to
differentiate between real and fraudulent photos; both networks are trained on the CelebA dataset. The proposed design uses
convolutional layers in the Discriminator and transposed convolutional (deconvolutional) layers in the Generator to improve
image quality and training stability. Binary cross-entropy loss, the Adam optimizer, and suitable data normalization are some of
the strategies used to ensure effective learning and better convergence. The results prove that DCGANs are capable of creating
realistic facial representations. To evaluate the model's efficacy, we use both qualitative and quantitative measures, such as
Frechet Inception Distance (FID). FID measures how similar the distributions of the real and generated pictures are to one
another. The results of this study demonstrate that DCGANs are very promising for applications in the entertainment, virtual
reality, data augmentation, and AI-powered creative content production industries.
In recent years, the rapid advancement of Generative Artificial Intelligence (GenAI) has transformed the landscape of digital content creation, enabling high-fidelity image synthesis across various fields including healthcare, art, and computer vision. However, the proliferation of synthesized images has introduced critical challenges, specifically the need to distinguish real physical imagery from synthetic, AI-generated counterfeits. This research paper presents a comprehensive, end-to-end framework that addresses both generative synthesis and discriminative detection under hardware-constrained (CPU-only) environments. We implement two synthesis methodologies like a Deep Convolutional Generative Adversarial Network (DCGAN) and a Convolutional Variational Autoencoder (ConvVAE) - trained on real image distributions to generate synthetic data. Concurrently, we present a compact Convolutional Neural Network (MiniCNNClassifier) designed to detect and classify images as real or fake. The framework is validated using a balanced dataset of 60,000 images (30,000 real and 30,000 synthetic). Our preprocessing pipeline ensures uniform size and resolution across heterogeneous inputs. Experimental results demonstrate that the MiniCNNClassifier achieves an outstanding validation accuracy of 98.7% and a Precision of 99.5%, Recall of 97.8%, F1-score of 98.6% in detecting fake samples. Furthermore, we provide a qualitative and quantitative comparison of DCGAN and ConvVAE architectures, discussing trade-offs between training stability and sample fidelity. Finally, we host the models on an interactive Streamlit-based web interface to enable real-time generation and classification.
Manoj T S, K. Siddamallappa U, A. Jajur. J· World Journal of Advanced Re...· 0 citations
This paper presents a real-time image denoising and reconstruction system based on a compact Generative Adversarial Network (GAN) designed for embedded systems and edge computing. The proposed model employs a [ generator and a PatchGAN discriminator, trained using hybrid loss function that combines L1, perceptual, and adversarial terms to balance pixel and perceptual realism. Evaluations were conducted on multiple datasets, including DIV2K, BSD68, SIDD, DND, RENOIR, and PolyU at noise levels (σ = 15, 25, 50). Quantitatively, the proposed model achieves an average PSNR of 32.8 dB and an SSIM of 0.88, which is higher by more than 3 dB on average than those of existing models, including DnCNN, FFDNet, and RIDNet. The ANOVA, Wilcoxon, and Friedman tests revealed the improvements were significant (p < 0.001). Moreover, the model can be inferred in less than 50 ms on the Jetson Nano and Raspberry Pi 4, confirming its applicability in real-time settings. At the same time, visual inspection and preference tests proved the high quality of perceptions and maintenance of texture. Overall, the proposed GAN is a powerful, efficient, and perception-guided real-time image recovery technique for practical computer vision applications.
Roa'a M. Al_airaji, H. Alrikabi, Rula Kamil et al.· Karbala International Journa...· 0 citations
In computer vision and pattern recognition tasks, deep learning models are widely used, especially in face recognition systems. Even with their excellent performance, these models are still susceptible to a variety of adversarial manipulations, such as blur, additive noise, translation, flipping, scaling, rotation, and changes in illumination. Furthermore, some architectures might experience optimization problems like vanishing gradients, which would further impair the stability of the model. In order to provide robust face verification under adversarial attack, Optimized Deep Learning–based Adversarial Defense Mechanism (ODL-ADM) is proposed in this work. It projects adversarial samples into an immune feature space. A Learnable Convolutional Principal Component Network (LCPCN) is incorporated into the framework to create a representation space that is both discriminative and resistant to perturbations. Adversarially corrupted facial images are suppressed and reconstructed using a Stacked Attention-based Residual Generative Adversarial Network (SARGAN). Accurate identity recognition is achieved by an Improved Cross-Triple MobileNetV1 architecture after perturbation removal. Enhanced Fire Hawk Optimization (EFHO) is used for performance maximization and parameter tuning to further improve recognition performance. Following image reconstruction and adversarial perturbation removal, the suggested model achieves a 98% face recognition accuracy.
Sheilla Ann Bangoy Pacheco, Mahesh Goyani, Jayzel P. Bangoy et al.· ITEGAM- Journal of Engineeri...· 0 citations
A dual-domain adversarial training framework is presented for high-fidelity 4D facial makeup detection and generation under complex and variable conditions. This approach employs paired generator-discriminator architectures, each responsible for either makeup or non-makeup facial sequences, integrated via explicit cross-domain feature alignment modules. Using composite loss strategies such as knowledge and adversarial loss, the system enforces consistent and realistic generation on both fronts. Temporal regularization and temporal regularization are among these methods. The sequence learning layer and the cascaded 3D convolutional layers enhance spatiotemporal modeling to achieve robust feature abstraction in facial dynamics. At the same time, by improving dynamic data, the training distribution and system adaptability have been enhanced. The extensive validation of the large-scale 4D annotated facial dataset shows significant improvements in recognition accuracy, generation stability, and structural preservation compared to traditional and deep learning standards. Quantitative analysis shows consistent resilience to severe illumination changes, occlusion, and pose variation. Ablation experiments indicate that each module is necessary, and efficiency evaluations show that the method is scalable. The results indicate that developing domain-aware alignment and hybrid loss integration techniques is beneficial for effective facial analysis in both controlled and challenging environments. This study promotes the practical application of intelligent human-computer interaction, providing various technical solutions for realistic 4D facial recognition and modeling.
Huihui Yin, Yurui Guan· International Conference on...· 0 citations
Accurate image segmentation in medical and industrial domains remains challenging due to small object sizes, complex textures, and diverse defect morphologies. To address these limitations, we propose Dense121GAN, a conditional generative adversarial network (cGAN) that integrates a pre-trained DenseNet121 as a frozen encoder. The proposed architecture promotes effective feature reuse and stable information flow through residual skip connections. By leveraging transfer learning, Dense121GAN captures rich hierarchical representations and improves training stability, particularly in data-constrained settings. Extensive experiments were conducted using five-fold cross-validation across 16 heterogeneous datasets. Dense121GAN consistently outperformed UNet and ResNet-based generative adversarial networks (GANs) across multiple evaluation metrics, including Intersection over Union (IoU), Dice coefficient (Dice), Matthews Correlation Coefficient (MCC), Cohen’s Kappa, Structural Similarity Index Measure (SSIM), coefficient of determination (
R
2
), and Symmetric Mean Absolute Percentage Error (SMAPE). The model also demonstrated stable training behavior and mitigated common GAN failure modes such as mode collapse. In addition, Dense121GAN achieved faster convergence compared to the evaluated baseline architectures. These results highlight the effectiveness, robustness, and computational efficiency of Dense121GAN. The proposed framework provides a practical and versatile solution for both clinical diagnostic and industrial inspection applications.
Muhammed Davud· PeerJ Computer Science· 0 citations
Recent advancements in deep learning have enabled the generation of realistic images directly from natural language descriptions. This paper presents a semantic-aware framework for text-to-face image synthesis using a joint Bidirectional Long Short-Term Memory (BiLSTM) network and Generative Adversarial Network (GAN). The proposed approach simultaneously trains the text encoder and image generator, allowing effective learning of semantic relationships between textual attributes and facial features. Initially, input descriptions are transformed into meaningful vector representations using Bi-LSTM, which are then utilized by the GAN to synthesize high-quality facial images. Unlike conventional methods that rely on separately trained text encoders, the proposed end-to-end architecture improves semantic consistency and visual realism. The model is trained on the CelebA dataset with corresponding facial descriptions and evaluated using similarity and image quality measures. Experimental results demonstrate improved face generation accuracy and better preservation of facial attributes, making the framework suitable for applications in forensic investigations, digital character creation, intelligent human-computer interaction, and public safety systems.
Heena Anjum, Sk.Mahammadunnisa· American Journal of AI Cyber...· 0 citations