基于Vision Transformer Hashing的民族布艺图案哈希检索算法研究  

Research on Hash Retrieval Algorithm for Ethnic Fabric Patterns Based on Vision Transformer Hashing

作　　者：韩雨萌 宁涛 段晓东 高原 HAN Yu-meng;NING Tao;DUAN Xiao-dong;GAO Yuan(School of Computer Science and Engineering,Dalian Minzu University,Dalian Liaoning 116650,China,Dalian Minzu University,Dalian Liaoning 116650,China;Key Laboratory of Big Data Application Technology of the National Civil Affairs Commission,Dalian Minzu University,Dalian Liaoning 116650,China)

机构地区：[1]大连民族大学计算机科学与工程学院,辽宁大连116650 [2]大连民族大学大数据应用技术国家民委重点实验室,辽宁大连116650

出　　处：《大连民族大学学报》2023年第3期250-254,共5页Journal of Dalian Minzu University

基　　金：辽宁省社科联基金项目(2022lsljdybkt-014);辽宁省经济社会发展研究科研项目(2022lsljdybkt-014)。

摘　　要：针对民族布艺图案复杂多样、语义提取、图像识别与检索困难等问题,以蜡染图案和织锦图案为例,提出一种图像检索算法,以提高匹配和检索民族布艺图案的准确度。结合民族布艺图案领域知识将民族布艺图案图像进行预处理,使用VIT为主干网络在哈希检索算法框架下进行图像检索。该方法优化了深度哈希检索算法,通过自身的自注意力机制提升了提取图案深层语义特征的能力,提高了深度哈希算法检索民族布艺图案的速度和精度。实验结果表明:提出的方法最佳检索精度可以达到95.32%。Aiming at the problems of complex and diverse ethnic fabric patterns,semantic extraction,and image recognition and retrieval difficulties,taking batik patterns and brocade patterns as examples,an image retrieval algorithm is proposed to improve the accuracy of matching and retrieving of ethnic fabric patterns.Combining knowledge in the field of ethnic fabric patterns,ethnic fabric pattern images are preprocessed,and image retrieval is performed using VIT as the backbone network under the framework of a hash retrieval algorithm.This method optimizes the deep hash retrieval algorithm,improves the ability to extract deep semantic features of patterns through its own self attention mechanism,and raises the speed and accuracy of the deep hash algorithm in retrieving national fabric patterns.Experimental results show that the optimal retrieval accuracy of the proposed method can reach 95.32%.

关 键 词：图像检索 深度哈希检索 VIT 民族布艺图案检索 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]