结合自适应网格描述符和图像能量的图像哈希  

Image hashing combining adaptive grid descriptor and image energy

作　　者：孙非凡 赵琰 SUN Feifan;ZHAO Yan(College of Electronics and Information Engineering,Shanghai University of Electric Power,Shanghai 201306,China)

机构地区：[1]上海电力大学电子与信息工程学院,上海201306

出　　处：《西安电子科技大学学报》2024年第5期136-148,共13页Journal of Xidian University

基　　金：国家自然科学基金(61802250)。

摘　　要：传统的图像分割方法对图像进行非重叠分块,图像哈希算法只能对图像块内部像素进行特征提取,而简单线性迭代聚类超像素分割算法能在聚合相似像素的同时额外提取形状特征,故在此基础上提出一种结合自适应网格描述符和图像能量的图像哈希算法。首先,使用双线性插值和高斯低通滤波对输入图像进行预处理,并通过简单线性迭代聚类对预处理图像进行超像素分割,对分割得到的超像素区域应用自适应网格描述符提取形状特征;同时利用超像素内部的像素具有相似亮度的特性,以亮度为计算对象求出各超像素的能量值作为图像的能量特征。最后,连接形状特征序列和能量特征序列,再利用密钥加密得到最终哈希序列。实验表明算法在鲁棒性和区别性之间达到了较好的平衡。算法的哈希运算时间为0.128 s,哈希长度为467 bits,具有较快的运算速度且哈希序列较为紧凑。分类性能方面,假阳性率为0时,真阳性率达到0.9999。拷贝检测方面,查全率和查准率均在95%以上。此外,与一些同类型算法相比,提出的算法在分类性能和拷贝检测方面也更具优势。Traditional image segmentation methods perform non overlapping segmentation on images.The image hashing algorithm can only extract features from the pixels inside the image block.However,the simple linear iterative clustering superpixel segmentation algorithm can extract additional shape features while aggregating similar pixels.Therefore,this paper proposes an image hashing algorithm that combines an adaptive grid descriptor and image energy based on simple linear iterative clustering.First,the input image is preprocessed through bilinear interpolation and Gaussian low-pass filtering.Then,simple linear iterative clustering is used to perform superpixel segmentation on the preprocessed image,and an adaptive grid descriptor is applied to extract shape features from the superpixels.Second,pixels within the superpixels have similar brightness characteristics,so the energy values of each superpixel region are calculated based on brightness as the energy feature of the image.Finally,the shape feature sequence and the energy feature sequence are connected.The final hash sequence is obtained by encrypting the connected sequence.Experiments show that the proposed algorithm achieves a good balance between robustness and discrimination.The average operation time and hash length of the algorithm are 0.128 s and 467 bits respectively,which leads to a fast operational speed and a compact hash sequence.In terms of classification performance,when the false positive rate is 0,the true positive rate reaches 0.9999.In terms of copy detection,both recall and precision are above 95%.In addition,compared with some similar algorithms,the proposed algorithm also has advantages in classification performance and copy detection.

关 键 词：图像处理 超像素 网格描述符 图像能量 拷贝检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]