一种运用图像力场特性的图像拷贝检测算法  

Image Copy Detection Algorithm by Image Force Field Mapping

作　　者：张冰[1] 昝程[1] 

机构地区：[1]深圳大学计算机与软件学院,深圳518060

出　　处：《计算机科学》2014年第11期306-308,共3页Computer Science

基　　金：深圳市科技研发资金基础研究基金(JC201005280459A)资助

摘　　要：将物理学中用引力场来描述分子间相互作用的思想应用于图像处理,图像力场的计算过程中将像素看作物理学中的粒子,像素的灰度值作为粒子的质量。分析表明,图像力场能够更直观地表示图像的离散信息且与图像本身是同构的。基于图像力场特性,提出了一个称为时钟算法的图像拷贝检测算法,其用表示时钟的时分秒针的3个线段来描述图像力场的主要特性并以此来表示图像的特征。实验结果表明,时钟算法可以通过较小的开销抵抗多种攻击,对噪声类失真与几何失真都有较好的鲁棒性,且可以通过设定检测策略分析遭受的攻击类型与强度,同时检测算法的查全率与查准率可保持在较高水平。The concept of force field to describe intermolecular interaction in physics is adapted to image processing.Image force field is designed as taking pixels for particles and grey scale for mass.The force field of an image models the discrete information of the image in a more intuitive way and is shown to be isomorphic to the image itself.Based on the image force field,an image copy detection algorithm called Clock algorithm was presented.In Clock algorithm,features of an image force field map are represented by three lines with the same endpoint like hour,minute and second hands of a clock,hence its name.The Clock algorithm was tested with images in an image database.The experiment results show that it can detect both noise-like distortions and geometric distortions with very high accuracy.In addition,the type and intensity of the distortion can also be determined,maintaining high rate at both recall and precision ratio in the same time.

关 键 词：图像拷贝检测算法 基于内容的拷贝检测 力场 噪声及几何畸变 分子计算 

分 类 号：TP317.4[自动化与计算机技术—计算机软件与理论]