基于深度学习的SIFT图像检索算法  

SIFT Image Retrieval Algorithm Based on Deep Learning

作　　者：苏勇刚 高茂庭[2] SU Yong-Gang;GAO Mao-Ting(Changzhou Institute of Industry Technology,Changzhou 213164,China;Shanghai Maritime University,Shanghai 201306,China)

机构地区：[1]常州工业职业技术学院,常州213164 [2]上海海事大学,上海201306

出　　处：《计算机系统应用》2020年第9期164-170,共7页Computer Systems & Applications

基　　金：国家自然科学基金(61202022)。

摘　　要：深度学习作为一个新的机器学习方向,被应用到计算机视觉领域上成效显著.为了解决分布式的尺度不变特征转换(Scale-Invariant Feature Transform,SIFT)算法效率低和图像特征提取粗糙问题,提出一种基于深度学习的SIFT图像检索算法.算法思想:在Spark平台上,利用深度卷积神经网络(Convolutional Neural Network,CNN)模型进行SIFT特征抽取,再利用支持向量机(Support Vector Machine,SVM)对图像库进行无监督聚类,然后再利用自适应的图像特征度量来对检索结果进行重排序,以改善用户体验.在Corel图像集上的实验结果显示,与传统SIFT算法相比,基于深度学习的SIFT图像检索算法的查准率和查全率大约提升了30个百分点,检索效率得到了提高,检索结果图像排序也得到了优化.Deep learning is a new filed in machine learning research,and to apply it to computer vision achieves effective result.To solve the problem that the traditional Scale-Invariant Feature Transform algorithm(SIFT)has low efficiency and extracts image features roughly,A SIFT image retrieval algorithm based on deep learning is proposed.The algorithm idea is that on the Spark platform,a deep Convolutional Neural Network(CNN)model is used for SIFT feature extraction,and Support Vector Machine(SVM)is utilized for unsupervised clustering of image library,then the adaptive image feature measures are used to re-sort the search results to improve the user experience.The experiment results on the Corel image set show that compared with the traditional SIFT algorithm,the precision and recall rate of the SIFT image retrieval algorithm based on deep learning is increased by about 30 percentage points and the retrieval efficiency is improved,the resulting image order is also optimized.

关 键 词：卷积神经网络(CNN) 深度学习 图像检索 重排序 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]