基于栈式卷积自编码的哈希图像检索研究

Research on Hash image retrieval based on stacked convolution auto-encoders

针对哈希结合深度学习在应用于图像检索时,存在准确率低下和耗时问题,提出一种基于栈似卷积自编码的哈希图像检索方法。文中提出以栈式卷积自编提取图像深层特征和哈希编码相结合的方法对图像进行检索。本方法首先使用栈似卷积自编码模型预训练图像数据,得到图像特征的深层表示,再采用哈希编码策略将这些深层表示进行编码,最后根据汉明距离的相似性来进行检索。实验部分,本文在两个大型数据集Fashion-MNIST和CIFAR-10上进行大量实验,并与现有哈希图像检索技术相比,证明了该方法在保证准确率的同时,提高了效率,实现了图像的快速精确检索。

Aiming at the low accuracy and time-consuming problem of hash-based deep learning when applied to image retrieval,a hash image retrieval method based on stacked convolution auto-encoders is proposed.In this paper,the method of stacked convolution auto-encoders image deep feature and hash coding is proposed to retrieve the image.The method firstly uses the stacked convolution auto-encoders model to pre-train the image data,obtains the deep representation of the image features,and then encodes these deep representations by hash coding strategy,and finally searches according to the similarity of Hamming distance.In the experimental part,this paper conducts a large number of experiments on two large data sets,Fashion-MNIST and CIFAR-10,and compared with the existing hash image retrieval technology,it proves that the method improves the efficiency and achieves the efficiency while ensuring the accuracy.It achieves fast and accurate retrieval of images.