基于梯度和语义特征融合的商标图像相似性研究  

Research on Trademark Iimage Similarity Bbased on Ggradient and Semantic Feature fusion

作　　者：马无锡[1] MA WuXi(Zhejiang Industry&Vocational College,Wenzhou 325003,China)

机构地区：[1]浙江工贸职业技术学院,浙江温州325003

出　　处：《浙江工贸职业技术学院学报》2024年第2期67-70,83,共5页Journal of Zhejiang Industry & Trade Vocational College

基　　金：2022年温州市科技局基础性科研项目“一种用于图形商标生成模型及检索的研究”(S20220041)。

摘　　要：图像检索在众多领域中起着不可替代的作用,也是计算机视觉领域重要的研究内容之一,特别是如何在海量的图像中快速检索到准确的图像。从图像特征提取入手,针对商标图像边缘信息丰富的特点,利用CLIP预训练模型,提出了一种CLIP特征和HOG特征相融合的特征提取方法,并且在两份数据集上进行对比实验,实验结果显示该融合特征优于使用单一特征的方法。基于该方法的有效性,开发了一套简易的商标图像相似性检索系统,可以快速从商标图像库中检索出和输入商标相似性高的图片。Image retrieval plays an irreplaceable role in various fields and is also one of the important research topics in the field of computer vision,especially concerning how to quickly retrieve accurate images from massive datasets.Focusing on the plentiful edge information in trademark images,this paper starts with the extraction of image features and then proposes a new feature extraction method that combines CLIP features with HOG features,which is based on the CLIP pre-training model.Retrieval experiments are conducted on two different datasets,and the results show that the methods with fused features outperform those using a single feature.Leveraging the effectiveness of this approach,a simple system for trademark image similarity retrieval is developed,which enables the rapidly-retrieved images from the trademark image database to be highly similar with the input trademark ones.

关 键 词：CLIP HOG 特征融合 图像相似性 商标图像 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]