【作者】 王威；

【导师】 沃焱；

【作者基本信息】 华南理工大学 ， 计算机应用技术， 2012， 硕士

【摘要】 随着图像处理技术的不断发展，抗几何攻击的特征在图像处理领域里扮演着越来越重要的角色，它被广泛应用于模式识别，3D建模，图像配准等方向。图像特征一般指那些相比于它周边邻域的有突出地方的元素，提取图像特征一般从灰度，颜色，纹理三方面进行提取工作。目前的特征提取技术能抵抗滤波，压缩等操作，但对于几何变换像缩放，旋转这些不影响图像内容的几何操作却没有好的方法，往往提取出与源图像完全不同的特征。而实际上，放缩，旋转这些对图像内容并无改变。本文旨在寻求一种能够容忍几何变换的特征提取算法，试图通过从图像中提取特征点的方法，提取出图像内容的特征，为局部水印嵌入提供参考点，最终构造出符合要求的水印嵌入区域。目前，图像特征提取主要有以下几种方式：点提取，区域提取；其中点提取因为其提取数量多，具有内容代表性而被大量研究使用。所以在本文中提及的算法均为基于点提取的，以下为本论文主要工作：(1)介绍目前主流的特征点提取算法包括SIFT，Harris，Harris-Laplace，DRSIF等。(2)提出一种基于Harris的分块特征提取算法。(3)根据SIFT的水印区域构造算法，提出一种基于Harris-Laplace的水印区域构造算法，并与基于SIFT的算法进行了实验比较，效果较前者优秀。(4)根据DRSIF特征提取算法的缺点，提出了一种新的基于DRSIF与显著区域的水印区域构造算法。(5)将本文提出的算法与基于SIFT的算法从水印区域的准确率，算法耗费时间等方面进行了实验比较，并进行了总结。从本文的实验对比可以看出，作者提出的两种基于特征点的水印区域构造算法均取得了较好的准确率，对于几何攻击有较好的鲁棒性。更多还原

【Abstract】 With the continuous development of image processing technology, the characteristics ofanti-geometric attacks plays an increasingly important role in the field of image processing.Itis widely used in pattern recognition,3D modeling,image registration and other fields.Image features generally refers to those prominent elements compared to its surroundingneighborhood. Image features are generally extracted from grayscale, color, texture. Thecurrent feature extraction are resistant to operation such as filtering, compression, but to thosegeometric transformation like rotation, scaling often extract completely differentcharacteristics compare to the characteristics extracted from the source image. In fact, scaling,rotating does not change image content. This paper seeks a feature extraction algorithm whichtolerate the geometric transformation. The algorithm try to extract the characteristics of theimage content through extracting feature points, and provide a reference point for the localwatermark embedding, and ultimately constructed the watermark embedding area.At present, the image feature extraction has the following ways: points extraction,region extraction; the point of extraction has been the mainstream because of the largenumber of features and representative of the image content. Algorithms mentioned in thispaper are based on the point of extraction, the following is the work of the paper:(1) Introduced the representative feature point extraction algorithm such as SIFT, Harris,Harris-Laplace, DRSIF.(2) Proposed an extraction algorithm based on the Harris.(3) Base on the watermark area construction algorithm based on SIFT, proposed a algorithmbase on Harris-Laplace for constructing watermark embedding area and the effect is betterthan the former.(4) According to the shortcomings of the DRSIF feature extraction algorithm, a new algorithmbased on DRSIF and salient region selection are proposed.(5) Compared the proposed algorithms with the algorithm based on SIFT from the accuracyand algorithm execution time, and made a summary. It can be seen from the experimental two watermark region construction algorithmproposed by this paper obtain a better accuracy for geometric attacks.更多还原