摘要
基于神经网络的非线性映射特性,提出了一种数字图像置乱的新方法。将图像输入到一个随机设置初始权值和阈值的BP神经网络,其输出即为置乱图像。只需训练一个从置乱图像到原始图像的神经网络,就可在恢复过程中将置乱图像输入到训练好的神经网络。实验结果显示,该算法对图像加密具有良好的效果,对噪声、JPEG压缩和剪切等攻击具有较好的抵抗能力。
A new method digital image scrambling based on the nonlinear mapping characteristics of neural networks is proposed. If an image is input into a BP neural network with arbitrary initial weights and thresholds, then the output is the scrambled image. A neural network, whose input is the scrambled image and the output is the original image, is needed. The scrambled image is input into the trained neural network when recovering. Experimental results show that this algorithm is effective for the image en- eryption, and is robust to the noise, JEPG compression, cropping attacks etc.
出处
《武汉理工大学学报（信息与管理工程版）》
CAS
2007年第10期5-9,共5页
Journal of Wuhan University of Technology：Information & Management Engineering
基金
陕西省自然科学基金资助项目(2004F10)
陕西省教育厅专项基金资助项目(05JK260)
关键词
图像置乱
神经网络
攻击
image scrambling
neural network
attack
