基于深度学习的图像翻拍检测

Recaptured Image Detection Based on Depth Learning

随着图像拍摄以及显示技术的发展,图像翻拍质量越来越高,这类图像可能会用作非法途径而不能被认证系统所识别。针对此问题,设计了一种基于深度学习的图像翻拍检测方法,该方法首先分析翻拍图像和真实图像之间的灰度变化,然后将其作为深度信念网络的输入逐层学习抽象的区分特征,最终在网络顶层实现图像的分类。实验结果表明,应用深度学习的方法能正确检测出翻拍图像。

With the development of image capture and display technology, high resolution photos may be used illegal ways because not recognized by the authentication system. Focus on this problem, the method is proposed based on deep learning. This method rooted on gray scale variation between real image and image produced by real image uses gray characteristic as depth belief networks input to learn more abstract feature that can achieve image classification at the top of the network. The experimental results show that the method can correctly detect image produced by real image.