Published Paper Details:

The increasing use of biometric authentication and online video content has made security more vulnerable to spoofing attempts and deepfake manipulations. Spoofing relies on presenting fake visuals, such as photographs or replayed clips, while deepfakes employ advanced generative models to produce highly convincing synthetic videos. Both threats compromise trust in identity verification systems and the authenticity of digital media. Existing approaches typically handle these problems in isolation, which limits their effectiveness in real-world applications. This work introduces Secure Vision, an integrated framework that combines MobileNet for real-time spoofing detection with ResNeXt for deepfake identification. MobileNet is optimized for speed and detects liveness indicators like eye blinks and facial micro-movements, while ResNeXt captures detailed features and uncovers inconsistencies in manipulated videos. The system is trained on benchmark datasets including ASVspoof, FaceForensics++, and the Deepfake Detection Challenge, improving its adaptability to different attack types. Results show that this dual-branch approach strengthens protection against both spoofing and deepfake threats, offering a practical solution for secure access control, online identity checks, and content authenticity verification.