文章摘要
李惠,方潜生,苏亮亮.基于多特征非线性融合的草图-图像检索方法[J].安徽建筑大学学报,2022,30():
基于多特征非线性融合的草图-图像检索方法
Sketch-image Retrieval Method Based on Multi-feature Non-linear Fusion
投稿时间:2022-05-10  修订日期:2022-05-22
DOI:
中文关键词: 草图-自然图像检索  手工特征  深度学习  特征融合
英文关键词: sketch-natural image retrieval, manual feature, deep learning, feature fusion
基金项目:
作者单位邮编
李惠 安徽建筑大学 智能建筑与建筑节能安徽省重点实验室 230022
方潜生 安徽建筑大学 智能建筑与建筑节能安徽省重点实验室 
苏亮亮* 安徽建筑大学 智能建筑与建筑节能安徽省重点实验室 230022
摘要点击次数: 1830
全文下载次数: 633
中文摘要:
      草图-图像检索属于计算机视觉领域的跨域检索问题,由于草图的简单性和抽象性,导致与自然图像的域间差异过大,至今其检索精度仍无法满足现实需求。于是本文提出了一种基于多特征非线性融合的草图-图像检索方法,通过非线性融合契合草图的形状特征和具有语义特性的深度特征。该融合策略结合了两类特征的优点,使之既能有效刻画边缘轮廓信息,又能兼顾语义信息。最后在两个公开数据集(Flickr15k和TU-Berlin)上进行了广泛实验,其结果显示本文提出的特征融合方法得到的平均检索精度优于其他融合方法和基于单一特征的方法,另外本文提出的融合策略易于扩展到其他多特征融合情况。
英文摘要:
      Sketch-image retrieval is a cross-domain retrieval problem in the field of computer vision. Because of the simplicity and abstractness of sketch, the retrieval precision of sketch is still not satisfied. In this paper, a sketch-image retrieval method based on multi-feature non-linear fusion is proposed. By non-linear fusion, the shape feature and depth feature of sketch are fitted. The fusion strategy combines the advantages of two kinds of features, which can not only describe the edge contour information effectively, but also give attention to the semantic information. Finally, extensive experiments on two public data sets (Flickr15k and TU-Berlin) show that the mean average precision of the proposed feature fusion method is better than other fusion methods and single feature-based methods. In addition, the fusion strategy proposed in this paper can be easily extended to other multi-feature fusion cases.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮