结合纹理复杂度与 BEMD 的灰度水印图像算法

本文提出并实现了一种结合纹理复杂度和二维经验模态分解(BEMD)的灰度水印图像算法。该算法首先对 Arnold 变换的水印图像进行 BEMD 分解，得到不同尺度的内蕴模态函数及余量信息。其次，对宿主图像分块进行纹理复杂度的计算，选择纹理复杂度高的区域作为嵌入位置；为了使水印图像的内蕴模态函数能与宿主图像更好地融合，对宿主图像进行了与水印图像相同筛分条件的 BEMD 分解。最后，将水印信息重复地嵌入到宿主图像预先选好的位置中，再对嵌入水印信息后的内蕴模态函数及余量重建得到嵌入水印后的图像。水印提取则为水印嵌入的逆过程。通过对不同纹理程度的宿主图像进行实验，得到嵌入水印后图像的峰值信噪比均在 40 dB 以上，面对 8 种常见攻击时提取水印的 NC 值均在 0.95 以上。与现有算法进行对比，该算法在大比例剪切、噪声攻击、图像滤波和 JPEG 压缩攻击上表现出色，且结果整体优于对比算法。

Grayscale watermarking algorithm via BEMD and texture complexity

This paper presented a grayscale watermarking algorithm combining texture complexity and bi-dimensional empirical mode decomposition (BEMD). Firstly, the watermark image after Arnold transform was decomposed by BEMD to obtain the intrinsic modal function (IMF) and the residual information of different scales. Then, the texture complexity of the host image after segmentation was calculated, and the region of high texture complexity was selected as the embedding location. In order to better fuse the IMF of the watermark image with the host image, BEMD decomposition was performed on the host image under the same sieving conditions as those for the watermark image. Finally, the watermark information was repeatedly embedded into the pre-selected location of the host image, and then the image embedded with watermarks was reconstructed by the IMF and the residual. Watermark extraction was the inverse process of watermark embedding. The experiments on host images at different texture levels revealed that the peak signal-to-noise ratios of images embedded with watermarks were all above 40 dB, and the normalized coefficients values of watermark extraction all exceeded 0.95 in the face of eight common attacks. Compared with the existing algorithms, the proposed algorithm performs well in large scale shearing, noise attack, image filtering, and JPEG compression attack, superior to the algorithms under comparison.