Jul 2026· International Journal for Research in Applied Science and Engineering Technology· Vol 14, pp. 1641-1647· 0 citations
TL;DR
BlinkNet, an explainable deepfake-detection framework that examines the spatial appearance and temporal kinematics of eye blinks, indicates that ocular dynamics can complement spatial evidence while improving efficiency and interpretability.
Abstract
The increasing realism of synthetic facial videos has reduced the reliability of detectors that depend only on visible
artifacts in isolated frames. This paper presents BlinkNet, an explainable deepfake-detection framework that examines the
spatial appearance and temporal kinematics of eye blinks. The system detects a face, localizes 68 facial landmarks, extracts
normalized ocular crops, and computes the Eye Aspect Ratio (EAR) for each frame. Overlapping sequences of 20 frames are
processed by a dual-stream Temporal-Spatial Physiological Blink Anomaly Network (TPBAN). A lightweight MobileNetV2
encoder models local visual inconsistencies, while a bidirectional gated recurrent unit models the forward and backward
dynamics of eyelid motion. Temporal attention assigns a relevance weight to every frame and supports frame-level anomaly
visualization. Training uses a multi-task objective for authenticity classification and blink-phase recognition, together with a
class-weighted binary cross-entropy term to address the imbalance between genuine and manipulated sequences. On the
FaceForensics++ c23 test partition, BlinkNet obtained 83.16% accuracy, 87.31% ROC-AUC, 96.02% average precision, and a
21.08% equal error rate. The manipulated class achieved 0.90 precision and 0.88 recall. The implementation processed video at
approximately 52 frames per second on a consumer laptop GPU and was integrated into a Flask-based forensic dashboard.
These results indicate that ocular dynamics can complement spatial evidence while improving efficiency and interpretability.
This work proposes a spatial-temporal model with two key components: one targeting artifacts within individual frames and the other analyzing inconsistencies across consecutive frames, both leveraging a bidirectional long short-term memory (Bi-LSTM) mechanism.
Ahmed Tammam, Hatem Abdel-Kader, Amira Abdelatey et al.· Signal, Image and Video Proc...· 0 citations
A detection framework is proposed that extracts per-video rPPG wave- forms via RhythmFormer and trains a suite of lightweight classifiers to distinguish real from synthesized physiologi- cal signals and shows that detec- tion difficulty is strongly method-dependent.
DeepFakes pose significant risks to digital security by enabling realistic facial manipulations that can evade conventional visual inspection. This study presents an attention-enhanced EfficientNet-B7 framework with a Custom Soft Spatial Attention (CSSA) module designed to localize manipulation-sensitive facial regions, including eye boundaries, mouth contours, blending boundaries, and skin-texture discontinuities. Evaluation is conducted on three benchmarks, a balanced Kaggle image subset containing 20,000 facial images, Celeb-DF (v2) with 50,000 sampled frames, and FaceForensics++ Low Quality with 40,000 sampled frames. The proposed model achieves 93.28% held-out test accuracy on the Kaggle subset, representing a 4.94 percentage-point improvement over standard EfficientNet-B7 and a 1.83 percentage-point improvement over an EfficientNet-B7+CBAM variant. Cross-benchmark evaluation yields 87.45% accuracy on Celeb-DF (v2) and 87.31% on FaceForensics++ (LQ). The model achieves a held-out test accuracy of 93.28% and a ROC AUC of 0.97. With 69.8M parameters, 15.0 GFLOPs, a 266 MB model footprint, and 150 ms single-image inference latency under the stated hardware configuration, the framework provides a transparent accuracy-efficiency trade-off for operational screening. Ablation testing confirms a 4.94 percentage-point contribution from CSSA, and LIME-based qualitative analysis supports interpretation of the learned decision regions.
Kislay Raj, Raja Vavekanand, Aditya Singh· Journal of Computer Virology...· 0 citations
The rapid development of artificial intelligence has led to the emergence of deepfakes, which pose serious threats to information security and public trust in digital media. This study develops a facial deepfake detection system that integrates YOLOv11 for face detection and the Xception architecture for classifying real and manipulated faces. YOLOv11 successfully localized all facial regions in the tested dataset with high confidence scores. The Xception model achieved a testing accuracy of 90.10%, with a Recall of 97.11% for the Fake class and an AUC of 0.98. Visual explanation using Grad-CAM showed that the model focused on critical areas such as the forehead, temples, and face boundaries to detect manipulation artifacts. The system was implemented as a desktop application named "Snap Detector" and passed black-box testing. However, the average processing speed of 6.26 FPS on an NVIDIA T4 GPU indicates that further optimization is needed for real-time performance.
Fachril Akbar, Kurniawati· JOURNAL OF APPLIED INFORMATI...· 0 citations
A novel method for video deepfake detection that assimilates the Pelican Optimization algorithm with a DL model jointly named as Pelican Attention Stacked Bidirectional Long-Short Term Memory (PAttSBiL), aimed at improving recognition accuracy and efficacy is presented.
D. R. Agrawal, Farha Haneef· Multimedia tools and applica...· 0 citations
The existing emotion detection systems are either audio or video centric. While such approaches have proven effective in controlled environments like a lab or a studio, these systems fail to perform under real-world conditions such as low light, audio interference, mispronunciation, and other environmental factors. The objective of our research is to design a quad-modal emotion detection system that combines four sub-models (static facial images, dynamic facial expressions, speech tone, and natural language processing [NLP] of the spoken words) to design a robust model that can perform under real-world conditions. We propose a dynamic weighting algorithm that biases the model towards a certain modality if the conditions are favorable (e.g. switch to audio-only if the video is obstructed) using cross-modal multi-head attention. We optimise the system to run on a CPU (due to the lack of a CUDA-enabled NVIDIA GPU) by using an asynchronous state buffer for the heavy NLP pipeline to maintain a high enough video frame rate. Our model achieved an accuracy of 61.21% (95% confidence interval [CI]: 58.77%-63.65%) with a mean area under the curve of 0.90 (95% CI: 0.88-0.92) across six emotion classes on real-world data (RAVDESS, SAMM, CREMA-D) during training and testing while maintaining performance under visual and audio deprivation conditions. Our architecture, while forgoing the precision of a studio-lit, well-audio-recorded environment, is able to sacrifice some accuracy for the robustness required to function in a real-world human-computer interaction environment.
