基于联合特征与中心方向信息的图像哈希算法

为了提高哈希技术对旋转操作的识别能力,提出了全局-局部联合特征耦合中心方向信息估计的图像哈希认证技术.首先,引入2D线性插值技术,对输入的图像进行预处理,使其对任意的缩放操作都具有固定尺寸的哈希序列;然后,将预处理图像转变为HSV彩色空间,借助二维离散小波变换(Discrete Wave Transform,DWT)处理V分量,利用其低频系数形成二次图像;再引入奇异值分解(Singular Value Decomposition,SVD)处理二次图像,提取其全局特征,将其作为第一个中间哈希序列;基于Fourier机制,借助残差方法,确定图像的显著区域,获取其位置与纹理的局部特征,作为第二个中间哈希序列;随后,引入Radon变换,通过计算图像的中心方向信息,将其与2个中间哈希序列组合,形成过渡哈希数组;借助Logistic映射,定义动态引擎参数,从而设计了分段异扩散技术,对过渡哈希数组进行加密,输出最终的哈希序列;最后,通过估算原始哈希序列与待检测哈希序列的Hamming距离,将其与用户阈值进行比较,完成图像认证.实验结果显示:与当前的图像哈希技术相比,所提算法具有更高的鲁棒性与安全性,对旋转攻击能力具有更好的识别能力.

Image Hashing Algorithm Based on Joint Feature and Central Orientation Information

In order to improve the authentication ability of hashing algorithms for rotary attacks, an image hashing authentication algorithm based on global-local joint feature and central orientation information was designed. Firstly, the 2D linear interpolation technique was introduced to preprocess the input image so that it had a fixed size hash sequence for any scaling operation. Subsequently, the preprocessed image was transformed into HSV color space, and a secondary image was formed based on low frequency coefficients by using the DWT (discrete wave transform) to transform the V components. Then the global features were extracted as the first intermediate hash sequence by introducing the SVD (singular value decomposition) to process the secondary images. Then the saliency region of the image was extracted based on the Fourier transform and spectral residual method to obtain the local features of its position and texture, which were taken as the second intermediate hash sequences. Then the central orientation information of the image was computed based on Radon transform to combine it with two intermediate hash sequences to form a transition hash array. The piecewise heterogeneous diffusion technique was designed by using the logistic map to define the dynamic engine parameters for encrypting the transition hash array and outputting the final hash sequence. Finally, the image authentication was completed by comparing the Hamming distance between the original hash sequence and the hash sequence to be detected. The results of an experiment showed that compared with the current image hashing technique, this algorithm had higher robustness and security, and better recognition ability for rotation attacks.