Abstract: A secure and robust image Hash method is proposed based on wavelet transform, pseudo-random block and singular value decomposition. After wavelet transform sub-band is divided into overlapping sub-image blocks, so that the main information of image can be obtained, and the smaller block of tampering with regional information may be obtained by the overlapping block. By pseudo-random dividing and pseudo-random scrambling, Hash cannot be forged without the key, so security of Hash is enhanced. Experimental results show that the proposed method is robust to content-preserving modifications such as JPEG compression, median filtering, additive noise, distortion, re-scaling and the slight geometry transformation, and it is sensitive to different contents and the operation of localized tampering.

Key words: image Hash, Singular Value Decomposition（SVD）, feature extraction, image authentication

摘要： 通过小波变换、伪随机分块以及奇异值分解，提出了一种安全的、鲁棒性较强的图像Hash方法。小波变换后将低频子带伪随机分割成相互重叠的子图像块，这样可以更好地获得图像的主要信息，块的重叠可以获得比分块更小的篡改区域的信息；伪随机分块和伪随机置乱，使得攻击者在不知道密钥的情况下无法伪造Hash，增强了Hash的安全性。实验表明，该方法对JPEG压缩、中值滤波、添加噪音、扭曲、图像缩放以及轻微的几何变换等内容保持不变的操作具有较好的鲁棒性，对于内容不同和局部篡改的操作具有一定的敏感性。

关键词: 图像Hash, 奇异值分解, 特征提取, 图像认证