一种基于CNN的图片隐写方法  

A Method of Image Steganography Based on CNN

作　　者：任志敏[1] Ren Zhimin(College of Mechanical Engineering,Changzhou Vocational Institute of Textile and Garment,Changzhou Jiangsu,213164,China)

机构地区：[1]常州纺织服装职业技术学院机电学院,江苏常州213164

出　　处：《山西电子技术》2023年第6期80-83,89,共5页Shanxi Electronic Technology

基　　金：江苏省大学生创新创业训练计划(202112807001Y);2020年常州纺织服装职业技术学院优秀青年骨干教师项目(czfzrc2020001)。

摘　　要：传统的信息隐藏方法通常是将有效载荷数据直接写入到宿主图片对应的二进制位,这有可能导致宿主图片的失真。为了解决这个问题,提出了一个基于卷积神经网络CNN的编码器和解码器架构嵌入有效载荷数据生成Stego图片,并从中恢复有效载荷图片。同时引入了一个新的损失函数,实现了训练集、验证集的低损失率,保证了编码器和解码器的训练。在CIFAR-100数据集上的实验结果表明,该方法在增大嵌入载入容量的同时,Stego图片和提取载荷图片质量都表现非常稳定。Traditional information hiding methods usually write the payload data directly into the corresponding binary bits of the host image,which may cause distortion of the host image.To solve this problem,a convolutional neural network(CNN)based encoder and decoder architecture is proposed to embed payload data to generate Stego images and recover payload images from them.At the same time,a new loss function is introduced to achieve a low loss rate for the training set and validation set,that ensuring the training of the encoder and decoder.The experimental results on the CIFAR-100 data set show that the method increases the embedding loading capacity,and the quality of the Stego image and the extraction load image are very stable.

关 键 词：隐写术 卷积神经网络 图片 编码 解码 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]