摘要
基于草图的自然图像检索是计算机视觉领域的重要研究内容之一。针对草图纹理、颜色等信息的缺乏导致检索精度的不足,尝试将有效刻画边缘轮廓信息的传统手工特征与有效反映语义信息的深度特征进行融合。首先分别对草图和自然图像进行下采样,并提取自然图像的边缘图;接着提取所有图像的HOG特征与深度特征;然后将两种特征进行融合得到最终的特征表示;最后通过特征间的相似性度量在公开数据集上进行检索实验,其结果显示,上述方法优于使用单一特征的方法。
Sketch-based natural image retrieval is one of the most important research fields in computer vision.Aiming at the deficiency of retrieval accuracy caused by the lack of information such as texture and color of sketch,this paper attempts to fuse the traditional manual features that effectively depict edge contour information with the depth features that effectively reflect semantic information.Firstly,the sketches and natural images were respectively down-sampled,and the edge map of natural images was extracted.Then HOG features and depth features of all images were extracted;The two features were fused to obtain the final feature representation later.Finally,the retrieval experiment was carried out on the public data set through the similarity measure between features,and the results show that the proposed method is superior to the method using single feature.
作者
方潜生
李惠
苏亮亮
杨亚龙
FANG Qian-sheng;LI Hui;SU Liang-liang;YANG Ya-long(Anhui Province Key Laboratory of Intelligent Building and Building Energy Saving,Anhui Jianzhu University,Hehei Anhui 230601,China;AnhuiInstitute of Strategic Studyon Carbon Dioxide Emissions Peakand Carbon Neutralityin Urban-Rural Development Hefei Anhui 230601,China;School of Electronic and Information Engineering,Anhui Jianzhu University,Hehei Anhui 230022,China)
出处
《计算机仿真》
北大核心
2023年第8期258-263,共6页
Computer Simulation
基金
国家自然科学基金(62001004)
安徽省高校省级自然科学研究项目(KJ2019A0768)
安徽省重点研究与开发计划项目(202104a07020017)
安徽建筑大学校级科研项目(2020XMK04)。
关键词
草图检索
多尺度边缘提取
深度学习
特征融合
Sketch-based image retrieval
Multiscale canny
Deep learning
Feature fusion
