摘要
针对残差网络直接应用于服装图像检索时,检索出的图片常存在款式相似但颜色差异较大的问题,提出一种融合传统颜色特征和基于ResNet50的深度特征的服装图像检索方法。充分利用残差网络在图像特征提取上的有效性和层次性,融合服装图片的深层网络特征和颜色特征作为最终特征向量,进行相似性度量,得到检索结果。实验结果表明,该方法能结合图片多种特征信息,提高了其检索准确率,具较强的鲁棒性和实用性,优于其他主流的检索方法。
When the residual network is directly applied to clothing image retrieval,the images retrieved often have similar styles but different color differences.Therefore,we propose a clothing image retrieval method combining traditional color features and depth features based on ResNet50.It made full use of the effectiveness and hierarchy of residual network in image feature extraction,and integrated deep network features and color features of clothing images as the final feature vector to measure the similarity and get the retrieval results.The experimental results show that the method can combine multiple feature information of images,and improve the retrieval accuracy.It has strong robustness and practicability,which is superior to other mainstream retrieval methods.
作者
侯媛媛
何儒汉
刘军平
Hou Yuanyuan;He Ruhan;Liu Junping(Engineering Research Center of Hubei Province for Clothing Information,Wuhan 430200,Hubei,China)
出处
《计算机应用与软件》
北大核心
2020年第10期194-199,共6页
Computer Applications and Software
基金
国家自然科学基金面上项目(61170093)。
关键词
服装图像检索
残差网络
深层网络特征
颜色特征
Clothing image retrieval
Residual network
Deep network feature
Color feature
