摘要
图像哈希是近年来颇受关注的一个研究热点,但现有的图像哈希算法普遍存在对几何攻击(主要包括尺度、旋转、剪切)鲁棒性不足的缺点,不能满足很多实际应用的需要。针对上述问题,提出了对几何攻击具有强鲁棒性的SIH图像哈希算法。本算法基于在图像匹配等领域得到广泛应用的SIFT算子,通过对SIFT特征向量进行有针对性的筛选和压缩、基于特征向量分布质心的量化生成图像摘要。为适应图像摘要构造的特性,设计了基于广义集合距的匹配算法来衡量图像摘要间的距离。在公开图像库上的实验结果表明,本算法对几何攻击和非几何攻击的鲁棒性均优于对比算法,可广泛服务于图像识别/认证类型的应用。
Image hashing has been a research hotspot these years. Since most existing image hashing algorithms are not robust enough to geometric attacks, such as scale, rotation and cropping attacks, they cannot meet the requirements of many practical applications. To solve this problem, SIH algorithm is proposed in this paper, which is a new image hashing algorithm robust to geometric attacks. Our algorithm is based on the SIFT operator, which has been used in image matching field. The image digest is constructed through filtering and compressing the SIFT feature vec- tor, which is followed by the quantization based on the centroid of feature vector. To fit the characteristic of the structure of image digest, a generalized set distance based matching operation is designed in this paper, which is not the hamming distance based matching commonly used in existing image hashing algorithms. Experimental results on the public image database demonstrate that our algorithm outperforms the typical comparative algorithms both on geometric attacks and non-geometric attacks. The proposed algorithm can be widely used in image identification/authen- tication related applications.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2011年第9期2024-2028,共5页
Chinese Journal of Scientific Instrument
