基于频谱特征的图像旋转角度盲检测算法

Image Rotation Angle Blind Detecting Algorithm Based Spectrum Character

抗几何攻击的盲检测数字水印的难点是图像旋转角度的盲检测.通过对图像频谱的分析,发现了图像边界蕴藏在幅度中的特征谱脊,设计出实际可行的频谱特征提取滤波器,其中特征锐化可以消除非轴线谱脊的干扰,最后对提取的二值特征图像采用霍夫变换测算谱脊的角度,即图像旋转角度模90°的余值.实验证明,角度测算不需要原始图像,具有计算量小、精度高的特点,对同时遭受几何攻击和滤波、加噪、有损压缩等处理都能有效测算出图像模90°的旋转角度.

It＇s difficult to detect blindly the image rotation angle in blind watermarking under geometry attack. By analyzing image spectrum, the character of spectrum amplitude ridge contained in image borderline is discovered. The practical distinguishing character filter is designed while character sharp filter can avoid the interfere whose amplitude ridge isn＇t axial. The angle of spectrum amplitude ridge in black-white character image can be calculated by Hough transform, that is the value of image rotation angle model π/2. The experiment result proves the proposed algorithm has light computational load and high precision without original image. It can endure the attacks of geometry attack, low-pass filter, noise additional, lossy compression at the same time.