基于统计特征的鲁棒图像拷贝检测技术

Robust image copy detection based on statistical features

抗几何攻击的拷贝检测技术是图像拷贝检测领域中的一个重要问题,现有的图像拷贝检测方法中,多数对类似噪声的图像失真比较鲁棒,而对于几何类失真则比较脆弱。因此,提出一种鲁棒图像拷贝检测技术,利用图像高斯过滤后的低频部分中的统计特征(直方图形状和均值)进行拷贝检测。实验结果表明该方法能较好地抵抗不改变内容的几何失真和图像处理攻击。

Copy detection resistance to geometric attacks is an important issue in the image copy detection community. Currently, most of the image copy detection methods can successfully resist the noise-like distortion and are quite fragile to geometric distortion. In this paper, the authors presented an image copy detection scheme by the use of two statistical features （the histogram shape and the mean） in low-frequency component of images after Gaussian filtering. Experimental results demonstrate that the proposed scheme can resist content-preserving geometric deformations and image processing attacks.