摘要
提出了一种新颖的基于尺度不变特征变换(SIFT)和主成分分析(PCA)的感知哈希方法。SIFT特征在通常的图像处理中具有很强的稳定性,并具有尺度和旋转不变性,通过对哈希生成两阶段框架的详细分析,SIFT算法用来提取图像的局部特征点,PCA用来对特征数据的信息压缩。每个特征点的PCA基的叠加构成图像哈希,在叠加中采用了伪随机处理,增强了算法安全性,图像之间的相似度通过哈希的归一化相关值来确定。实验分析表明该方法对各种复杂攻击,如图像旋转、光照变化、图像滤波等具有较好的稳健性,对比基于非负矩阵分解的图像哈希方法在图像识别应用中具有更好的性能。
This paper presents a novel perceptual hashing method based on scale invariant feature transform（SIFT） and principal component analysis（PCA）.SIFT features are invariant to image scaling and rotation,and stable to common image processing.Through the detailed analysis of two-stage framework of generating hash,SIFT is used to capture the local features of image.PCA is used to compress features information.Final hash is generated by summing PCA basis of each key point.The method uses pesduo-randomly processing for enhancing security of the algorithm.The similarity of images is determined by hash normalization correlation.Test results indicate that the proposed method is robust to various types of attacks such as image rotation,illumination change and filtering,etc.It is superior in image identification compared with the image hashing using non-negative matrix factorization.
出处
《电路与系统学报》
北大核心
2013年第1期274-278,228,共6页
Journal of Circuits and Systems
基金
国家自然科学基金资助项目(61001201)
