结合卷积神经网络与哈希编码的图像检索方法

Image retrieval method combining convolution neural network and Hash coding

为了提高图像检索的精度与速度,提出一种卷积神经网络与哈希方法结合的图像检索算法。该方法在深度残差网络的基础上构建了一个网络模型,将随机选取成对的图像(相似/不相似)作为训练输入,使用曼哈顿距离作为损失函数,并添加了一个二值约束正则项,促使训练好的网络输出为类二值码,再将类二值码阈值化为二值码,最后用于图像检索。在Caltech256数据集和MNIST数据集上的实验结果显示,文中方法的检索性能优于其他现有方法。

In order to improve the precision and speed of image retrieval,an image retrieval algorithm combining convolutional neural network and Hash method is proposed in this paper.On the basis of the depth residual network,a network model is built with the proposed method,which can select pairs of images(similar/dissimilar)randomly as a training input.The Manhattan distance is taken as a loss function,and a binary constraint regularization is added to prompt the trained network output as approximate binary code.Then the approximate binary code threshold is converted into binary code for image retrieval.The results of experimental on Caltech256 data set and MNIST data set show that the retrieval performance of this method is better than other existing methods.