The traditional artificial visual features based image hash retrieval method do not necessarily preserve the semantic similarities of images,resulting in poor retrieval performance issue,an efficient convolutional neural network based hash coding retrieval method was proposed.The original F7 layer of AlexNet network was removed,a new whole connection layer with 48 nodes and using the sigmoid function as activation function was added.The improved network model was used to fine-tune the target data sets and a threshold of 0.5 was set to binarize the activate value of the new added layer to get the hash binary code. The hash coding was used to retrieval.Results of experiments on CIFAR-10 data sets show that the performance of the proposed method increases 30% compared to CNNH+,it also superior to other methods on the MINST data sets,verifying the effective-ness of the proposed method for large-scale image retrieval.%为解决传统的基于人工视觉特征的图像哈希检索方法不一定能保留图像语义相似性,导致检索性能不好的问题,提出一个高效的基于卷积神经网络的哈希编码检索方法.去掉AlexNet网络原来的F7层,加入新的节点为48且激活函数为sigmoid函数的全连接层,采用该改进网络模型针对目标数据集进行微调,阈值设置为0.5,二值化新加入层的激活值得到哈希二进制编码,采用哈希编码进行检索.在CIFAR-10数据集上的实验结果表明,该方法与CNNH+相比性能提升了30%,在MINST数据集上也优于其它方法,验证了该方法对于大规模图像检索的有效性.
展开▼
