Image Hashing for Tamper Detection with Multiview Embedding and Perceptual Saliency

Table 2

Comparison with deep learning based methods for image forensics.

Method	Main technique	Application	Hashing
Chen [15]	Filter layer that output median filtering residual	Median Filtering detection	No
Qian [16]	Customized CNN model	Steganalysis	No
Bayar [17]	New convolutional layer to learn manipulation detection features	Image manipulation (i.e. median filtering, gaussian blurring, additive white Gaussian noise, eesampling) detection	No
Bondi [18]	Clustering of camera-based CNN features	Tampering Detection and Localization	No
Yarlagadda [19]	GAN and one-class classifier	Satellite image forgery detection and Localization	No
D’Avino [20]	Autoencoder with recurrent neural networks	Video forgery detection	No
Tuama [21]	A layer of preprocessing is added to the CNN model	Camera model identification	No
Bondi [22]	data-driven algorithm based on convolutional neural networks	Camera model identification	No
Proposed	Multiview feature, perceptual saliency, semi-supervised hashing	Image tamper detection	Yes