基于三元组与变分自编码器的图像检索算法  

Image Retrieval Algorithm Based on Triplet and Variational Autoencoders

作　　者：杨奎河[1] 刘怡[1] YNAG Kuihe;LIU Yi(Hebei University of Science and Technology,Hebei Shijiazhuang,050018,China)

机构地区：[1]河北科技大学,河北石家庄050018

出　　处：《长江信息通信》2023年第4期82-85,共4页Changjiang Information & Communications

摘　　要：基于内容的图像检索是计算机视觉领域的重要分支之一,其允许用户输入一张查询图像,通过特征匹配查找具有相同或相似内容的其它图像。在本文中,提出了一种新的图像检索框架,称为三重变分自编码器。该框架利用变分自编码器将输入的三元组嵌入潜在空间,并利用三元组学习类内相似性和类间差异性。同时,考虑到三元组的选择对有效训练三重损失的重要性,文章提出了一种三元组选择方法,在每个训练批次中选择与锚定图像最相似的正面例子和最相似的负面例子,同时保证这个选定的锚负对比锚正对距离大一个裕度的范围,以此提高三重损失的学习性能。实验结果表明,所提出的方法优于现有方法。Content-based image retrieval is one of the important branches in the field of computer vision.It allows users to input a query image and find other images with the same or similar content through feature matching.In this paper,we propose a new image retrieval framework called triple variational autoencoder.The framework uses the variational autoencoder to embed the input triplets into the potential space,and uses the triplets to learn the intra-class similarity and inter-class differences.At the same time,considering the importance of triplet selection for effective training of triple loss,this paper proposes a triplet selection method.In each training batch,select the most similar positive example and the most similar negative example with the anchor image,and at the same time ensure that the selected anchor negative is a margin larger than the distance between the anchor positive and negative,so as to improve the learning performance of triple loss.The experimental results show that the proposed method is superior to the existing methods.

关 键 词：图像检索 三联体网络 变分自编码器 二进制哈希码 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]