Scale change and rotation invariant digital image watermarking method
Scale change and rotation invariant digital image watermarking method
复制标题
尺度变化和旋转不变的数字图像水印方法
DOI：
10.1002/scj.20670
复制
发表时间：
2007
期刊：
影响因子：
--
通讯作者：
N. Hamada
中科院分区：
文献类型：
--
作者：
Mizuki Tone;N. Hamada
近年来，数字图像水印作为数字图像版权保护的一种方法受到了广泛的关注。由于图像用户根据不同的应用需要进行图像的放大或缩小等失真处理，因此需要具有抗几何失真能力的数字图像水印。为了能够从具有几何失真的图像中检测到水印，需要对像素相对位置的变化进行补偿。本文提出了一种利用特征点进行补偿，然后利用频率分量进行水印嵌入和检测的方法。特征点由Harris-Affine检测器获得。首先，从图像中提取特征点周围的椭球体，每个椭球体归一化为一个圆，并将水印嵌入到该频率分量中。然后通过在检测过程中以相同的方式提取椭球体并将椭球体归一化成圆来执行几何失真补偿以检测水印。使用特征点进行补偿消除了为原始图像和定位嵌入标记的需要，从而使补偿诸如比例和旋转等仿射失真成为可能。此外，由于为每个单个特征点确定一个椭圆嵌入区域，因此即使在几个特征点由于失真而检测失败的情况下，也可以正确地提取水印插入区域。《Wiley期刊》，Inc.Syst Comp Jpn，38(10)：1-11,2007；在线发表在Wiley InterScience(www.intercience.wiley.com)上。DOI 10.1002/scj.20670
In recent years attention has focused on digital image watermarking as a method for copyright protection of digital images. Since users of images make distortions such as enlarging or reducing images depending on the application, digital image watermarking which is robust to geometric distortions has been demanded. Compensation for changes in the relative position of pixels is needed in order to allow watermarking to be detected from images with geometric distortions. In this paper we propose a method that uses feature points to perform compensations and then use frequency components to embed and detect watermarks. The feature points are obtained by a Harris‐Affine detector. At first, ellipsoids around a feature point are extracted from an image, each ellipsoid normalized to a circle, and a watermark embedded into that frequency component. Geometric distortion compensation is then performed by extracting ellipsoids in the same manner during detection and normalizing the ellipsoids into circles to detect watermarks. Using feature points for the compensation eliminates the need to embed markers for original images and positioning thereby making it possible to compensate for affine distortions such as scale and rotation. Furthermore, since one elliptical embedment region is determined for each single feature point, the watermark insertion region can be correctly extracted even if detection of several feature points fails due to distortion. © 2007 Wiley Periodicals, Inc. Syst Comp Jpn, 38(10): 1– 11, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/scj.20670