Cross-Domain Recaptured Document Detection with Texture and Reflectance Characteristics

Recapturing attack is a simple but effective anti-forensic operation for concealing the forgery trace in a digital document image. In this paper, we focus on detecting the recap-turing operation in a questioned certificate document image. By analyzing the printing techniques used in producing a certificate document, we identify that the texture and reflectance character-istics in the bronzing region can be employed as discriminative features for recaptured document detection. A generic framework based on convolutional neural network is employed to evaluate the effectiveness of these features. To study our approach, two certificate document image datasets including 132 genuine and 1464 recaptured document images are established. Experimental results show that the proposed framework outperforms the state-of-the-art scheme under different experimental protocols. The proposed framework has achieved on average 0.90 AVC under the most difficult experimental protocol, i.e., the cross-domain experiment with different documents.