基于尺度不变特征变换的快速图像拷贝检测

摘要

Focusing on the problems of low feature extraction speed and low matching efficiency of the traditional image copy detection algorithm based on Scale Invariant Feature Transform (SIFF) feature,a fast image copy detection algorithm based on location distribution and orientation distribution features of SIFT feature points was proposed.Firstly,the twodimensional location information of SIFT feature points was extracted.The number of feature points in each interval was counted with block statistics by calculating the distance and angle between each feature point and image center point.The binary hash sequence was generated to construct the first order robust feature according to the quantitative relationship.Then,the numbers of sub-interval feature points in all directions were counted with block statistics according to the one-dimensional direction distribution feature of feature points,and the secondary image feature was constructed according to the quantitative relationship.Finally,a cascade filter framework was used in the copy detection to make a judgement about whether the copy or not.The simulation experimental results show that,compared with the traditional copy detection algorithm which constructs the hash sequence based on the SIFT feature with 128-dimensional descriptor,the feature extraction time of the proposed algorithm is shortened to the original 1/20,and the matching time is also reduced by more than 1/2.Therefore the proposed algorithm meet the requirement of online copy detection.%针对传统基于尺度不变特征变换(SIFT)特征的图像拷贝检测算法特征提取速度慢、匹配效率不高的问题,提出了一种基于SIFT特征点位置分布与方向分布特征的快速图像拷贝检测算法.首先,提取SIFT特征点二维位置信息,通过计算各个特征点与图像中心点的距离、角度,分块统计各区间的特征点数量,依据数量关系量化生成二值哈希序列,构成一级鲁棒特征;然后,根据特征点一维方向分布特征分块统计各方向子区间特征点数量,依据数量关系构成二级图像特征;最后,拷贝检测时采用级联式过滤框架作出是否为拷贝的判断.仿真实验结果表明,与传统SIFT以128维特征描述子为基础构建哈希序列的图像拷贝检测算法相比,所提算法在保证鲁棒性与独特性不降低的同时,特征提取时间缩短为原来的1/20,匹配时间也缩短了1/2以上,可满足在线拷贝检测的需求.