当前图像哈希认证技术对旋转操作较为敏感, 稳健性与伪造检测率不高。为克服此类问题, 设计了基于改进的局部二值模式(LBP)算子与动态更新变换的紧凑图像哈希算法。引入线性插值技术, 对输入图像实现预处理, 改善哈希序列对尺度缩放的稳健性。利用Ring分割, 将插值图像转化成二次图像。考虑中心像素特性与周围像素的差异, 将Heaviside函数嵌入传统的LBP中, 形成了描述能力较强的H-LBP算子, 提取图像的抗旋转稳健特征。引入压缩感知, 对高维特征矢量完成降维, 输出紧凑的过渡哈希数组。利用混沌变换思想, 设计动态更新变换机制, 对过渡哈希数组进行加密, 得到图像哈希序列。最后, 利用汉明距离测算输入图像与待检测图像的哈希相似度, 通过优化认证阈值, 完成图像内容的真伪识别。实验结果显示, 与当前哈希算法相比, 提出的算法生成的哈希序列尺寸更小, 对旋转、噪声等操作具有更好的感知稳健性。

In order to overcome such defects as low robustness and low detection rate induced by the sensitivity of current hash image authentication algorithm to the rotation operation, the compact image hash algorithm based on an improved local binary pattern (LBP) operator and the dynamic update transform is proposed. The linear interpolation technique is introduced to preprocess the input image and improve the scaling robustness of the hash sequence. The Ring division is used to transform the interpolation image into the secondary image. A H-LBP operator with strong descriptive ability was designed when the Heaviside function is embedded into traditional LBP operator. Considering the difference between the center pixel characteristics and the adjacent pixels, the anti-rotation robustness features are extracted. The compression sensing is introduced to reduce the dimensions of high-dimensional feature vector, and a compact transition hash array is output. A dynamic update transform mechanism is designed with the chaotic transform idea to encrypt the transition hash array, and the image hash sequence is obtained. Finally, the hash similarity between the input image and the detected image is calculated based on Hamming distance to achieve the authenticity of the image through the optimized decision threshold. The experimental results show that the proposed algorithm generates a smaller hash sequence and has better perceptual robustness to rotation, noise and other operations than the current hash algorithms.