图像块的不可见性与鲁棒性均衡水印算法

发布时间：2018-04-10 10:21

选题：人类视觉系统(HVS)　视角：奇异值分解(SVD)　引自：《中国图象图形学报》2017年06期

【摘要】：目的为协调水印算法不可见性与鲁棒性之间的矛盾,提高水印算法抵抗几何攻击的能力,提出一种图像块的不可见性与鲁棒性均衡水印算法。方法将宿主图像分成互不重叠的图像块,利用人类视觉系统的掩蔽特性对每个图像块的纹理特征和边缘特征进行分析,选择掩蔽性好的图像块作为嵌入子块。对嵌入子块作2级离散小波变换,将其低频子带进行奇异值分解,通过修改U矩阵第1列元素间的大小关系嵌入Arnold置乱后的水印信息。在水印提取前,对几何失真含水印图像利用图像尺度不变特征变换(SIFT)特征点的坐标关系和尺度特征进行几何校正,恢复水印的同步性。结果对标准灰度图像进行实验,含水印图像的峰值信噪比都可以达到44 d B以上。对含水印图像进行常规攻击和几何攻击,提取出的水印图像与原始水印图像的归一化互相关系数大部分都能达到0.99以上,说明该算法不仅具有良好的不可见性,对常见攻击和几何攻击都具有较强的鲁棒性。结论选择掩蔽性好的图像块作为水印嵌入位置能够充分保证水印算法的不可见性,特别是水印提取前利用SIFT特征点具有旋转、缩放和平移不变性对几何失真含水印图像实现有效校正,提高了含水印图像抵抗几何攻击的能力,较好地协调水印算法不可见性与鲁棒性之间的矛盾。
[Abstract]:In order to reconcile the contradiction between the invisibility and robustness of the watermarking algorithm and improve the ability of the watermarking algorithm to resist geometric attacks, a novel watermarking algorithm is proposed to equalize the invisibility and robustness of image blocks.Methods the host image was divided into non-overlapping image blocks, and the texture features and edge features of each image block were analyzed by using the masking characteristics of human visual system, and the image blocks with good masking ability were selected as embedded sub-blocks.The embedded sub-block is transformed by 2-level discrete wavelet transform and the low-frequency subband is decomposed by singular value decomposition. The watermark information after Arnold scrambling is embedded by modifying the size relation between the elements in the first column of the U matrix.Before the watermark is extracted, the geometric correction of the coordinate relation and the scale feature of the image's SIFT feature points is carried out to restore the synchronization of the watermark.Results the peak signal-to-noise ratio (PSNR) of the standard grayscale image is above 44 dB.The normalized correlation between the extracted watermark image and the original watermark image can reach more than 0.99, which indicates that the algorithm is not only invisible, but also can be used to improve the performance of the watermark image.It is robust to common attacks and geometric attacks.Conclusion selecting the image block with good masking as the watermark embedding position can fully guarantee the invisibility of the watermarking algorithm, especially the rotation of the SIFT feature points before the watermark extraction.Scaling and translation invariance can effectively correct the geometric distortion watermark images, improve the ability of watermark images to resist geometric attacks, and better coordinate the contradiction between the invisibility and robustness of the watermarking algorithm.
【作者单位】：辽宁工程技术大学软件学院;
【基金】：国家自然科学基金项目(61540056,61401185)~~
【分类号】：TP309.7;TP391.41

【相似文献】