Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{197490,
        author = {Runal A. Parlewar and Mr. Om P. Pote and Ms. Aditi A. Sharma and Mr. Aditya S. Hurde},
        title = {Temporal-Spatial Deep Neural Framework for Robust DeepFake Face Detection in Video Sequences},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {6508-6523},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=197490},
        abstract = {A The sudden development of adversarial generative models that produce life-like face-swapped videos has made concerns over digital content authenticity, abuse of an individual's identity, and propagation of misinformation a very real issue. This paper introduces RecureFaceNet - a sophisticated deep learning framework designed for facial integrity detection of videos by jointly learning the spatial features and temporal motion patterns. The proposed model comes with a two-stream design, which has a temporal branch which captures the inconsistencies in motion of multiple frames and small temporal distortions that are typical to deepfakes, and a spatial branch which focuses on the detailed facial textures for identifying the synthesis-related artefacts. A special fusion module merges and harmonizes the learned embeddings from both branches, so the model is able to successfully distinguish between real and falsified facial sequences. To enhance generalization ability to unseen types of manipulations, reconstruction-regularization loss is imposed on real face sequences, which leads to tightly clustered representations of real dynamics and outlier regions for fake embeddings. Experiments on popular benchmark tests show that RecureFaceNet performs better in detection accuracy, temporal robustness and cross-manipulation generalization than previous state-of-the-art methods. The results also put more stress on the benefit of using temporal information that considerably improves the reliability of detection compared to methods based on single frames. Finally, this work demonstrates the potential of RecureFaceNet for real-time forensic video analysis, as well as future directions for developing more adversarially robust and practical deepfake detectors.},
        keywords = {Deepfake detection; video forensics; RecureFaceNet; temporal-spatial feature; convolutional neural networks (CNN); recurrent neural networks (RNN); generative adversarial networks (GAN); digital media integrity; feature fusion; adversarial robustness.},
        month = {April},
        }