Abstract: A semi-fragile watermark algorithm based on mean computation and coefficient classification is proposed. It divides the original image into blocks, and carries through mean computation on each block. The decimals of those coefficients derived from mean computation follow certain rule. The rule is used to classify those coefficients into two groups. It makes the high four bits of every pixel right shift, and derives zero-watermark sequence by comparing the quantities of coefficients in the two groups. The inconsistency between the quantities of coefficients in every group resulted from tamper is used to test its ability to tamper location. Experimental result shows that the algorithm has good transparency and enough robustness to resist the common image processing operations. It is able to locate the tampered areas and it has low probability of miss. It is able to resist collage attack.

Key words: mean computation, coefficient classification, semi-fragile watermark, zero-watermark, tamper location

摘要： 提出一种基于均值运算和系数分类的半脆弱水印算法。在空域对原始图像进行分块，根据每个子块均值运算所得矩阵系数小数部分的规律性将系数分为2类。将图像各像素的高4 bit向右移位，通过每一子块2类系数的个数比较产生零水印。利用篡改前后每类系数个数的不一致性检测算法的篡改定位能力。实验结果表明，该算法的不可见性和鲁棒性较好，能抵抗常规图像处理操作，且具有定位篡改的能力，漏警概率低，能抗拼贴攻击。

关键词: 均值运算, 系数分类, 半脆弱水印, 零水印, 篡改定位

CLC Number: 

• TN911.4