图像相似性改进算法  

Improved algorithm of image similarity

作　　者：马进[1] 郝宁宁 李红宇 MA Jin;HAO Ningning;LI Hongyu(Department of Automation,North China Electric Power University,Baoding Hebei 071003,China)

机构地区：[1]华北电力大学自动化系,河北保定071003

出　　处：《计算机应用》2022年第S01期287-291,共5页journal of Computer Applications

基　　金：华北电力大学教学改革项目(1300200226)。

摘　　要：针对图像相似性度量问题,提出一种图像相似性改进算法。首先,分别得到源图像和候选图像的三个向量,包括HSV颜色空间H分量的颜色直方图向量、尺度不变特征转换(SIFT)算法局部特征向量以及图像颜色矩向量;然后,采用向量拼接的方法融合上述三个向量,借助主成分分析对融合后的向量降维,生成新的特征向量;最后,计算源图像和候选图像特征向量之间的欧氏距离,进行特征点匹配和图像相似度计算。实验结果表明,所提算法不仅提高了颜色直方图算法对于颜色分布接近图像的识别精度,而且减少了SIFT算法对于平坦区域较多图像的误判,平均查准率较颜色直方图算法和SIFT算法分别提高了23.8和9.6个百分点。所提算法能够有效提高识别图像相似性的准确度,判别结果更符合人眼观察结果。To address the image similarity measurement problem,an improved image similarity algorithm was proposed.Firstly,three vectors of the source image and the candidate image were obtained respectively,including the color histogram vector based on the H(Hue)component of HSV(Hue-Saturation-Value)color space,the local feature vector based on Scale-Invariant Feature Transform(SIFT)algorithm and the image color moment vector.Secondly,the above three vectors were fused by vector stitching method,and the fused vector was dimension-reduced with the help of principal component analysis to generate the new feature vectors.Finally,the Euclidean distance between feature vectors of the source image and the candidate image was calculated for feature point matching and similarity calculation.Experimental results show that the improved algorithm not only improves the recognition accuracy of the color histogram algorithm for images with close color distribution,but also improves the misjudgment of the SIFT algorithm for images with many flat areas,and the average accuracy is increased by 23.8 and 9.6 percentage points compared with the color histogram algorithm and SIFT algorithm respectively.The proposed algorithm can effectively improve the recognition accuracy of image similarity,and the discrimination result is more consistent with the observation result of human eyes.

关 键 词：图像相似度 颜色直方图 特征点匹配 颜色矩 数据降维 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]