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{197126,
        author = {Dr. B. Suribabu and Tamanampudi Sai Jahnavi and Korukonda Gamya Sri and Gollakoti Jyothi and Saladi Lakshmi Kantham and Pappula Krishna Veni and Dr. Y. Venkat},
        title = {Hybrid Forensic-Neural Fusion for Deepfake Image Detection Using Multi-Signal Analysis},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {6009-6015},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=197126},
        abstract = {Fabricated facial images have emerged as a significant concern across journalism, legal proceedings, and digital identity verification. While existing detection systems perform effectively under controlled conditions, they often fail when evaluated on images generated by previously unseen manipulation techniques. This limitation reduces their reliability in real-world scenarios, where the source and method of manipulation are typically unknown.
This paper proposes a detection approach that integrates two complementary analysis paradigms that have traditionally been studied independently: deep neural inference and classical image forensics. Three forensic signals are extracted from each input, namely compression error patterns across multiple quality levels, discontinuities at JPEG block boundaries, and inconsistencies in noise variance. These features are combined with the output of a pretrained convolutional neural network using a weighted score fusion strategy. In addition, a conservative correction mechanism is introduced to reduce false positives, particularly in genuine images.
Experimental evaluation on the Face Forensics++ and Celeb-DF v2 datasets achieves an accuracy of 91.3% and an AUC-ROC of 0.957. When tested on previously unseen data, the proposed model exhibits an accuracy drop of only 7.7 percentage points, which is approximately half the degradation observed in comparable single-model approaches. The entire pipeline operates efficiently on a standard CPU without requiring GPU acceleration and is implemented as both a web-based application and a real-time webcam detection system.},
        keywords = {Deepfake Detection, Image Forensics, Error Level Analysis, Score Fusion, Convolutional Neural Networks, Face Forensics++, Celeb-DF, JPEG Artifact Analysis, Digital Media Integrity},
        month = {April},
        }