摘要
为了提高汽车图像检索的准确率,选择了一个合适的颜色空间和量化算法。首先本文自建了1000张汽车图像库,其次在HSV颜色空间提出了梯形高斯隶属度量化算法进行特征提取,最后通过欧式距离和加权距离进行相似度匹配。实验结果表明,该算法对汽车图像进行检索时的平均查准率要比梯形隶属度量化算法在欧氏距离进行相似度量时平均高出11.2个百分点;在加权距离进行相似度量时平均高出10.9个百分点。
In order to improve the accuracy of image retrieval, a suitable color space and quantization algorithm are selected. First, a 1000 criminal vehicle image library is built in this paper. Secondly, the trapezoid Gauss membership quantization algorithm is proposed for feature extraction in HSV color space. Finally, the similarity is matched by the Euclidean distance and the weighted distance. The results show that the average precision of the algorithm is 11.2 percentage points higher than that of the trapezoid subordination algorithm at Euclidean distance, and the average height of the weighted distance is 10.9 percentage points higher than that of the weighted distance.
引文
[1]张强,李嘉锋,卓力.车辆识别技术综述[J].北京工业大学学报,2018,44(3):382~392.
[2]Zhang S,Tian Q,Huang Q,Gao W,Rui Y.Cascade category-Aware visual search[J].IEEE Transactions on Image Processing(TIP),2014.
[3]Liu Z,Li H,Zhang L,Zhou W,Tian Q.Cross-indexing of binary SIFT codes for large-scale image search[J].IEEE Transactions on Image Processing(TIP),2014.
[4]Liu Z,Li H,Zhou W,Zhao R,Tian Q.Contextual hashing for large-scale image search[J].IEEE Transactions on Image Processing(TIP),2014,23(4):1606~1614.
[5]Xie L,Tian Q,Zhou W,Zhang B.Fast and accurate near-duplicate image search with affinity propagation on the image web[J].Computer Vision and Image Understanding(CVIU),Special Issue on Large Scale Multimedia Semantic Indexing,2014,124:31~41.
[6]Norvig P.2020 visions[J].Nature,2010,463:26~32.
[7]Cai X,Nie P,Huang H,Kamangar F.Heterogeneous image feature integration viamulti-modal spectral clustering[C]//CVPR,2011:1977~1984.
[8]束鑫.基于区域模糊直方图的图像检索[J].江苏科技大学学报(自然科学版),2007,32(4):234~239.
[9]常小红,濮黄生,见伟平.一种基于HSV颜色空间模糊量化的图像检索方法[J].医疗卫生装备,2011,32(4):19~22.
[10]刘忠伟,章毓晋.十种基于颜色特征图像检索算法的比较和分析[J].信号处理,2000,(1):79~84.