基于卷积神经网络和监督核哈希的图像检索方法

发布时间：2018-03-20 06:08

本文选题：深度学习　切入点：图像检索　出处：《电子学报》2017年01期 　论文类型：期刊论文

【摘要】：当前主流的图像检索方法采用的视觉特征,缺乏自主学习能力,导致其图像表达能力不强,此外,传统的特征索引方法检索效率较低,难以适用于大规模图像数据.针对这些问题,本文提出了一种基于卷积神经网络和监督核哈希的图像检索方法.首先,利用卷积神经网络的学习能力挖掘训练图像内容的内在隐含关系,提取图像深层特征,增强特征的视觉表达能力和区分性;然后,利用监督核哈希方法对高维图像深层特征进行监督学习,并将高维特征映射到低维汉明空间中,生成紧致的哈希码;最后,在低维汉明空间中完成对大规模图像数据的有效检索.在Image Net-1000和Caltech-256数据集上的实验结果表明,本文方法能够有效地增强图像特征的表达能力,提高图像检索效率,优于当前主流方法.
[Abstract]:The visual features used in the current mainstream image retrieval methods lack the ability of autonomous learning, which leads to the poor performance of their image expression. In addition, the traditional feature indexing methods have low retrieval efficiency. It is difficult to apply to large scale image data. Aiming at these problems, this paper proposes an image retrieval method based on convolution neural network and supervised kernel hash. Using the learning ability of the convolutional neural network to mine the implicit relation of the training image content, extract the deep feature of the image, enhance the visual expression ability and distinguishing ability of the feature. The supervised kernel hash method is used to monitor the deep features of high-dimensional images, and the high-dimensional features are mapped to low-dimensional hamming spaces to generate compact hash codes. The experimental results on Image Net-1000 and Caltech-256 datasets show that the proposed method can effectively enhance the expression ability of image features and improve the efficiency of image retrieval. Better than the current mainstream method.
【作者单位】：解放军信息工程大学信息系统工程学院;武警工程大学电子技术系;河南工程学院计算机学院;
【基金】：国家自然科学基金(No.60872142,No.61301232)
【分类号】：TP391.41;TP183

【相似文献】