对局部内容篡改敏感的感知图像散列

Perceptual image hashing sensitive to content modification

提出了一种基于图像内容和颜色分布的感知图像散列。先将图像尺寸规格化并分成小块,根据各块亮度矩阵的奇异值判断其是否属于复杂区域,由此得到复杂区分布索引表。计算各图像块Y分量的均值和R、G、B均值两两之差的最小值,构成表征亮度和颜色分布的特征向量,将它与复杂区索引组合并加密得到图像散列。实验结果表明,由此提取的图像散列对保持图像内容不变的JPEG压缩、平滑滤波、缩放等处理具有良好的稳健性,而对内容篡改敏感并能准确定位篡改部位。

A perceptual image hashing method based on the image content and color distribution was proposed. The image was scaled to a fixed size and divided into blocks. Each block was then classified semantically into complex or plain according to the ratio of the first two singular values of its luminance matrix so as to give a table of complexity indices. RGB components of each complex block were used to form a color-feature vector, which was combined with the complexity table and encrypted to produce the image hash. Experiments show that the hash is robust against content-preserving image manipulations such as JPEG compression, smoothing and re-scaling, and sensitive to tampering of the image content.