基于双线性模型的图像检索技术  

Image retrieval research based on bilinear model

作　　者：曾凡锋 胡胜达 付国涛[2] ZENG Fan-feng;HU Sheng-da;FU Guo-tao(College of Computer,North China University of Technology,Beijing 100144,China;Research and Development Center,Huang-Gong Machinery Group)

机构地区：[1]北方工业大学计算机学院,北京100144 [2]黄工机械集团有限公司技术研发中心,浙江温州325100

出　　处：《计算机工程与设计》2019年第11期3225-3229,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(61371142);北京市教委科技创新服务能力建设基金项目(PXM2017-014212-000002)

摘　　要：为尽可能保留图像的有效信息,解决在传统的基于深度哈希的图像检索方法中,单一的信息流网络结构设计可能引发的部分图像信息缺失的问题,提出一种采用双线性模型网络结构的图像检索方法。两条并行的信息流能够充分对图像信息进行筛选并保留图像的有效信息,在后续的卷积过程中这些有效的图像信息可以得到充分融合,实现改善图像检索效果的目的。CIFAR-10和MNIST数据集上的实验结果表明,该方法相较现有基于深度哈希的图像检索方法有更好的图像检索效果。To preserve the effective information of images as much as possible and to solve the problem of the loss of partial effective information of images caused by the network structure design with single information flow in the existing deep hash-based image retrieval methods,an image retrieval method using bilinear model network structure was proposed.Two parallel information flows effectively filtered the image pixels and preserved the effective information of images,and this information could be fully integrated in the subsequent convolution process to achieve the propose of improving the performance of image retrieval.Experimental results on the CIFAR-10 and MNIST datasets show that the proposed method has better image retrieval performance than the existing deep-hash based image retrieval method.

关 键 词：图像检索 深度哈希 双线性模型 卷积神经网络 深度学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]