基于视皮层全局感知特征的顽健图像散列方法

Robust image hashing method from global perception features of human visual cortex

提出了一种基于视皮层全局感知特征的感知散列方法,图像首先被低通滤波后缩放成预定尺寸,然后图像分割成依赖于密钥的重叠子图像块,每块根据人类视觉系统主视皮层的层次结构,抽取图像经视觉通道逐层处理后的方向轮廓响应,并与表面颜色信息联合形成视皮层的全局感知特征,这些特征包含了每个图像块的方向、颜色信息,使得相邻系数的关系在通常的图像处理下保持不变,利用这种不变性将所有图像块的特征信息量化、置乱后形成二值图像散列。实验表明提出的方法对JPEG压缩、图像滤波等内容保持操作具有较好的顽健性,同时具有检测恶意篡改的能力,不同图像之间具有很低的共谋概率。

Firstly, the image was filtered low-passly and re-scaled to fixed size. Then the image was divided into key-dependent overlapping blocks. According to the hierarchical structure of HVS primary visual cortex, directional contour responses were extracted through layer-by-layer processing of visual pathway, and global perception features could be obtained by associating with the surface color of each blocks. These features included image direction and color information. The relation of adjacent entries was invariant to ordinary image processing. It could be used to quantize and permute these feature to get the binary image hash. Test results indicate that the proposed method is robust against content-preserving modifications such as JPEG compression and filtering, at the same time, is capable of localizing image malicious tampering. Hashes of different images have very low collision probability.