Impact Factor

7.883 Call For Paper

Share on

Author(s):

Gayatri P. Nandavadekar | Shrenika V. Salunkhe
Keywords:

AI-generated Image Detection, Image Forensics, Multi-layer Feature Analysis, Heuristic Classification, Streamlit, Digital Media Authenticity, Synthetic Media Verification.
Abstract:
The Proliferation Of AI-generated Images Poses Pressing Threats To Information Integrity, Cybersecurity, And Public Trust. This Paper Presents DeepScan AI, A Multi-layer, Heuristic-driven Image Authenticity Verification System Engineered Without Reliance On Large-scale Labeled Training Datasets. The Proposed System Extracts And Fuses Seven Distinct Feature Channels—EXIF Metadata Integrity, Luminance Noise Distribution, Texture Complexity (variance-based), RGB Channel Entropy, Edge-gradient Regularity, Frequency-domain Artifacts, And Aspect-ratio Consistency—to Generate A Probabilistic Authenticity Score In The Range [0, 100]. Deployed As A Lightweight Web Application Using Python And Streamlit, DeepScan AI Achieves An Experimental Detection Accuracy Of 85.5% With A Mean Inference Latency Of 2–4 Seconds, Operating Entirely On-device To Preserve User Privacy. Unlike Cloud-dependent Or Deep-learning-heavy Alternatives, The System Is Zero-cost, Transparent, And Immediately Accessible To Non-technical Users. Results Demonstrate That Multi-feature Fusion Significantly Outperforms Single-channel Heuristic Baselines And Establishes A Practical Foundation For Further Deep-learning Augmentation.