Sketch based image retrieval based on abstract-level transform and convolutional neural networks(融合抽象层级变换和卷积神经网络的手绘图像检索方法)

Sketch based image retrieval based on abstract-level transform and convolutional neural networks(融合抽象层级变换和卷积神经网络的手绘图像检索方法)

针对人工设计的描述子(HOG、SIFT等）在基于手绘的图像检索(Sketch Based Image Retrieval，SBIR)领域的局限性，提出了一种融合抽象层级变换和卷积神经网络构建联合深度特征描述子的手绘图像检索方法.首先，提取常规图像的边缘概率图，在此基础上进行不同抽象层级的图像变换，将抽象层级变换图像输入到深度神经网络并提取不同隐层的输出向量，最后，联合不同隐层的输出向量作为手绘图像检索的特征描述子（即联合深度特征描述子）.在Flickr15k数据库上对本方法进行了实验验证，结果表明：融合抽象层级变换和联合深度特征描述子的检索效果相较HOG、SIFT等传统方法有显著提高.本方法从...

Full description

Bibliographic Details
Main Authors: LIUYujie(刘玉杰), PANGYunping(庞芸萍), LIZongmin(李宗民), LIHua(李华)
Format: Article
Language:zho
Published: Zhejiang University Press 2016-11-01
Series:Zhejiang Daxue xuebao. Lixue ban
Subjects:
手绘检索
卷积神经网络
边缘概率检测
抽象层级变换
联合深度特征
Online Access:https://doi.org/10.3785/j.issn.1008-9497.2016.06.005

Internet

https://doi.org/10.3785/j.issn.1008-9497.2016.06.005

Similar Items