提出一种基于视觉特性的图像摘要算法，增大人眼敏感的频域系数在计算图像Hash时的权重，使得图像Hash更好地体现视觉特征，并提高鲁棒性。将原始图像的分块DCT系数乘以若干由密钥控制生成的伪随机矩阵，再对计算的结果进行基于分块的Watson人眼视觉特性处理，最后进行量化判决产生固定长度的图像Hash序列。本算法比未采用视觉特性的算法相比，提高了对JPEG压缩和高斯滤波的鲁棒性。图像摘要序列由密钥控制生成，具有安全性。-Based on the visual characteristics of the image digest algorithm, increasing the human eye-sensitive frequency-domain coefficients in the calculation of the image when the weight of Hash, Hash makes images better reflect the visual characteristics, and improve robustness. Will block the original image multiplied by the number of DCT coefficients generated by the key control of pseudo-random matrix, then the results of calculation based on the sub-block of Watson HVS treatment, and finally quantify the judgments arising from fixed-length sequence of images Hash . Than the algorithm did not use the visual characteristics of the algorithm, improve the JPEG compression and Gaussian filtering robustness. Abstract image sequence generated by the key control, with security.