Multimodal feature fusion for video forgery detection

In this paper we propose a novel local feature analysis and feature level fusion technique for detecting tampering or forgery for facial-biometric based on-line access control scenarios. The local features are extracted by analysing facial image data in the chrominance colour space and hue-saturation colour space. A feature level fusion of local features consisting of hue and saturation gradients with global features obtained from principal component analysis showed that a significant improvement in performance can be achieved in detecting tampered or forged images from genuine images in low bandwidth online streaming video access control contexts. The performance evaluation of the proposed fusion technique for a multimodal facial video corpus showed that an equal error rate of less than 1% could be achieved with feature level fusion of local features and global features.