智能与分布计算实验室
  用于拷贝检测的图像特征提取算法研究
姓名 陈勇
论文答辩日期 2008.06.05
论文提交日期 2008.06.10
论文级别 硕士
中文题名 用于拷贝检测的图像特征提取算法研究
英文题名 Research on Image Feature Extraction Algorithm in Copy Detection
导师1 卢正鼎
导师2
中文关键词 拷贝检测;特征提取;兴趣点;信息熵
英文关键词 copy detect;feature extraction;interest points;entropy
中文文摘 随着因特网(Internet)的发展和高效数字化存储设备价格的不断降低,使得现在很容易的通过网络来复制、传输和分发数字图像内容,因此对于知识产权(IPR)的有效保护成为了一项十分重要的事情。而对数字媒体(包括图像、声音和视频)的拷贝进行检测,是保护知识产权的一项基本的要求。 随着拷贝检测技术的发展,针对拷贝检测系统的攻击已经取得高度发展。总体上说,针对拷贝检测系统的攻击可以分为两种类型:类似噪声的信号处理和几何失真。其中几何失真可以导致原始图像与其拷贝版本之间发生位置同步误差,因而会造成较高的误检率。目前,绝大多数拷贝检测算法对类似噪声的攻击具有较好的抵抗能力,而对几何失真类攻击十分脆弱。可见,抗几何失真问题是拷贝检测领域难点问题。由于基于全局的特征几乎不能抵抗任何几何失真,采用对几何失真具有较好抵抗能力的局部特征提取方法是提高拷贝检测鲁棒性的有效办法。 根据上述的问题,提出一种使用兴趣点的拷贝检测特征提取算法。将计算机视觉中的兴趣点技术应用到图像的特征提取中。通过兴趣点的检测对图像的视觉特征变化大的区域进行定位,并提取兴趣点周围局部区域的信息熵,然后通过信息熵的差异比较进行局部区域的相似性度量得到相匹配的局部特征区域。接下来使用兴趣点的位置比较确定局部区域的空间一致性状况,最终确定图像的拷贝检测。兴趣点描述了图像局部特征的空间信息,信息熵则描述了图像局部区域的亮度分布。二者结合,使得该特征提取方法对图像的几何变换具有很强的鲁棒性,同时也能抵抗大多数的信号处理攻击。 实验结果显示该方法是十分有效的,并且相对于图像全局特征,基于兴趣点的特征,对于旋转,裁减,翻转等几何攻击具有较好的鲁棒性。
英文文摘 The success of the Internet and cost-effective digital storage device has made it possible to replicate, transmit, and distribute digital content in an effortless way. Thus, the protection of intellectual property right (IPR) has become a crucial legal issue. Detecting copies of digital media (images, audio and video) is a basic requirement for IPR protection (or copyright protection). With the development of copy detection technologies, attacks against copy detecting systems have become more sophisticated. In general, the attacks on copy detecting systems can be categorized into noise-like signal processing and geometric distortions. While geometric distortions induce synchronization errors between the original and the copies and therefore can mislead the copy detector. Most of the previous methods have shown robustness against noise-like signal processing attacks, and only a few specialized copy detecting methods have addressed the geometric distortions. Due to a weak resistance of global feature against geometric distortions, introducing a local feature extraction which has a better resistance against geometric distortions is a efficacious way to increase robustness of copy detection. According to the problems described above, the paper gives a copy detection feature extracting method using interest points. The solution applies technology of interest points to image feature extracting. First, they localized relevant regions by detecting interest points in the image. Then several small regions around the interest point is located as an image patch. Entropy information features are extracted to describe each image patch. Finally, the similarity measurement can be done by comparing entropy information of neighboring region. At last detect spatial distribution by using the location of interest points to decide one image a copy or not. Interest points represent spatial details of local features and entropy information represents distribution of local luminance. Integrating both of interest points and entropy information make the method has a good robust to geometric attack and signal processing. Experiments show that this method is very effective by our experiment system of interest points based image copy detection and has better robust to geometric attacking than the methods based on global features.