利用分块相似系数构造感知图像Hash

提出一种基于图像分块相似系数的感知稳健图像Hash.先对图像预处理,再进行重叠分块,在密钥控制下,利用高斯低通滤波器生成伪随机参考图像块,分别计算每个分块与参考图像块的相关系数得到图像特征序列.依此将相邻两个分块特征值合并以缩短Hash长度,同时对压缩后的特征序列进行重排,进一步提高图像Hash的安全性.最后对归一化特征值进行量化,并运用Huffman方法对其编码,进一步压缩Hash长度.理论分析和实验结果表明,该图像Hash方法对JPEG压缩、适度的噪声干扰、水印嵌入、图像缩放以及高斯低通滤波等常见图像处理有较好的鲁棒性,能有效区分不同图像,冲突概率低,可用于图像篡改检测.

Perceptual Image Hashing Using Block Similarity

A robust and perceptual image hashing method is proposed based on similarity coefficients of image blocks.The image is firstly preprocessed,and then divided into overlapped blocks.A reference pattern in relation to a secret key is generated with a Gaussian low-pass filter.The correlation coefficient between the reference pattern and each image block is calculated.These coefficients are concatenated and an intermediate hash is obtained.To get a short image hash,every two coefficients corresponding to neighboring blocks are merged into a new coefficient.These new coefficients are permuted pseudo-randomly to produce a secure hash.Finally,these new coefficients are mapped to the interval [0,100],and Huffman coded.Theoretical analysis and experimental results show that the proposed method is robust to JPEG compress,additive noise,watermark embedding,re-scaling and Gaussian low-pass filtering,and it has very low collision probability.It can be applied to tamper detection.