基于RAN与深度哈希的图像检索方法研究

Research on image retrieval method based on RAN and deep hashing

针对基于卷积神经网络提取的图像特征不仅包含目标信息,还包括杂乱的背景信息这一问题,提出了一种基于残差注意力网络与深度哈希的算法,该算法通过残差注意力网络提取图像特征,输入到哈希层得到图像的二进制编码,通过对比待查询图像的哈希码与训练集中每一张图像的哈希码之间的汉明距离来检索图像,可实现端到端的训练和检索。在Flickr和NUS-WIDE数据集上的实验结果表明,与残差网络相比,所提方法在平均检索精度上大约提升了1.1%~2.7%,有较高的准确率且检索性能相对稳定。

Aiming at the problem that image features extracted by convolutional neural network contain target information and messy background information,an algorithm based on residual attention network and deep hashing is proposed.In this algorithm,features are extracted through residual attention network and input into the hash layer to obtain the binary code of the image,and then some imagesare retrieved by comparing the hamming distance between the hash code of the image to be searched and the hash code of each image in the training set,which can realize end⁃to⁃end training and retrieval.The experimental results on Flickr and NUS⁃WIDE data set show that compared with residual network,the method proposed in this paper improves the average retrieval accuracy by about 1.1%-2.7%,with high accuracy and relatively stable retrieval performance.