Channel Difference Based Regeneration Architecture for Fake Colorized Image Detection

Shruti S. Phutke, Subrahmanyam Murala

Abstract: Selling of counterfeit/fraud goods like publishing the fake images of the products is increasing now a days due to advancement of multimedia technology for image editing applications. This work aims to detect the fake colorized images with a novel observation of channel difference maps. Here, we propose a novel deep channel difference network for fake colorized image detection/classification. Initially, an effective channel difference map (CDM) based auto-encoder is proposed for image regeneration. Here, to get distinguishing edge and color information between real and fake colorized images, CDM is proposed. After effective regeneration, the abstract version of learned features can be used for classification. Thus, we have proposed the transfer learning based network for fake colorized image detection (FCID) with learned encoder features of regeneration network. To the best of our knowledge, this is the first work with CDM based auto-encoder for FCID. The performance of the proposed network is tested on benchmark datasets and compared with the existing state-of-the-art methods for FCID in terms of half total error rate (HTER). The experimental results demonstrate that the proposed network is superior than the state-of-the-art approaches for FCID.