基于四元极谐变换矩与显著特征的图像鲁棒哈希算法

ROBUST IMAGE HASHING ALGORITHM BASED ON QUATERNION POLAR HARMONIC TRANSFORM MOMENTS AND SALIENT FEATURES

为了增强哈希序列对各种几何变换攻击的鲁棒性,设计基于四元极谐变换矩与显著特征的图像鲁棒哈希算法。引入线性插值与自适应Wiener滤波器,实现初始图像的预处理;计算预处理图像的颜色向量角度,并基于Fourier变换,得到其对应的幅度信息,以获取两个不同的频谱;计算两个频谱的残差,确定图像中的局部显著性区域;通过LBP算子,提取显著特征;基于四元极谐变换(Quaternion Polar Harmonic Transform,QPHT),获取预处理图像的QPHT矩;联合显著特征与QPHT矩,形成过渡哈希数组。引入Logistic映射,定义加密函数,实现对过渡哈希数组的加密,输出最终的哈希序列,以增强其抗碰撞性能。测量源图像与可疑图像之间的哈希序列所对应的l_(2)范数距离,并将其与优化阈值比较,对图像的真实性做出判断。在多种几何变换攻击下完成测试,输出数据显示:较当前准确性较高的哈希方法而言,该算法具有更理想的鲁棒性与识别准确率。

In order to enhance the robustness of hash sequences against various geometric transformation attacks,a robust image hashing algorithm based on quaternion polar harmonic transformation moment and salient features is designed.Linear interpolation and adaptive Wiener filter were introduced to realize the pre-processing of the initial image.The amplitude information of preprocessed image was calculated based on Fourier transform to obtain two different spectrums.Then the residual of two spectra was calculated to determine the local saliency region in the image.And the salient features were extracted by using LBP operator.The QPHT moment of preprocessed image was obtained base on quaternion polar harmonic transformation.The transition hash array was formed by combining salient features with QPHT moments.Logistic map was introduced to define the encryption function for encrypting the transition hash array and outputting the final hash sequence,which enhanced its anti-collision performance.The l_(2)norm distance of the hash sequence between the source image and the suspicious image was measured,and the authenticity of the image was judged.The test under a variety of geometric transformation attacks is completed,and the output data shows that this algorithm has better robustness and recognition accuracy than the current Hash method with higher accuracy.