基于离散余弦变换与最优嵌入强度预测的鲁棒图像水印算法  

Robust Image Watermarking Algorithm Based on Discrete Cosine Transform and Optimal Embedding Strength Prediction

作　　者：滕春燕 杨德运 TENG Chun-yan;YANG De-yun(Department of Information Technology, Jiangsu United Vocational and Technical College Xuzhou Branch of Finance and Economics, Xuzhou Jiangsu 221008, China;School of Software, Shandong University, Ji'nan 250100, China;School of Information Science and Technology, Taishan College, Tai'an Shandong 271000, China)

机构地区：[1]江苏联合职业技术学院徐州财经分院信息技术系,江苏徐州221008 [2]山东大学软件学院,济南250100 [3]泰山学院信息科学技术学院,山东泰安271000

出　　处：《西南师范大学学报（自然科学版）》2021年第1期50-58,共9页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：国家自然科学基金项目(61379015);山东省自然科学基金项目(ZR2011FM004);江苏省自然科学基金项目(BD2012129);江苏省高校“青蓝工程”资助项目(苏教师[2014]23号).

摘　　要：为了解决当前图像水印方案难以兼顾系统的抗几何攻击能力与视觉隐秘性的问题,提出了基于离散余弦变换与最优嵌入强度预测的鲁棒图像水印算法.把宿主目标划分为若干个8×8的非重叠子块;引入离散余弦变换DCT(Discrete Cosine Transform)处理每个子块,输出相应的低频与高频子带;从低频子带中选择一个系数作为参考系数,借助JPEG量化表,选择与参考系数相接近的4个不同系数;从选择的系数中确定出一个与参考系数最为接近的系数,存储其位置,视为密钥,并找出幅度值最大的系数;利用不同攻击类型的鲁棒性来构建人工蜂群算法的适应度函数,通过训练样本来获取最优嵌入强度值;借助水印融合原则,把密钥信息植入到参考系数与幅度值最大的系数中,得到水印结果;最后,根据其存储位置,提出水印复原方案,完成水印数据的检测.测试数据显示:与已有的水印方案相比,所提技术具有更好的水印效果,在多种几何攻击下,均具备更优的视觉隐秘性与鲁棒性.In order to solve the problem as impossible to both the geometric attack ability and visual concealment of the system in current image watermarking scheme,a robust image watermarking algorithm based on discrete cosine transform and optimal embedding strength prediction has beenproposed.Firstly,the host image was divided into several 8×8 non-overlapping sub-blocks,and the embedded watermark was divided into sub-watermark with the same number of host image.The discrete cosine transform was applied to deal with each sub block for outputting the corresponding low frequency and high frequency subbands.A coefficient was selected from the low frequency sub-band as the reference coefficient,and four different coefficients close to the reference coefficient were selected with the help of the JPEG quantization table.From these four different coefficients,the closest coefficients to the reference coefficients were determined which stored as the key,and the maximum amplitude coefficient were found out.The artificial bee colony algorithm was introduced to train the training samples for output a set of data by using the robustness of different attack types to construct its fitness function for predicting the optimal embedding strength value.The watermark embedding method was designed to hide the watermark content into the maximum amplitude coefficient and reference coefficient for outputting the watermark image.Finally,according to its storage location,a watermark extraction method was established to recover the watermark information.Test data show that,compared with existing watermarking schemes,this watermarking scheme has better watermarking effect,which has better visual concealment and robustness under various geometric attacks.

关 键 词：图像水印 离散余弦变换 人工蜂群 嵌入强度预测 参考系数 适应度函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]