Abstract: Aiming at the retrieval of images, a method based on Hash coding and Convolutional Neural Network（CNN） is proposed. The idea is to add a Hash layer to the CNN, and a coarse-to-fine search strategy is used to retrieval similar images. Firstly, the coarse-level search is operated according to the Hash coding to obtain the same or similar images to form a pool of m images. Then the Euclidean distances between the high-level semantic features of the m images and query image are calculated for fine-level search, thus achieving the ultimate retrieval purpose. The proposed method takes the loss of Hash layer as one of the optimization goals, and two features are combined for image retrieval, which makes up the shortcomings of time consuming and memory using of the existing methods. The results show that the proposed method outperforms the state-of-art algorithms on printed fabric and CIFAR-10 datasets.

Key words: image retrieval, Convolutional Neural Network（CNN）, Hash coding, hierarchical search

摘要： 针对图像检索，提出一种基于哈希编码和卷积神经网络的方法。主要是在卷积神经网络（CNN）中加入哈希层，采用由粗到精的分级检索策略，根据学习到的哈希码进行粗检索得到与查询图像相同或相似的[m]幅图像构成图像池，计算池内图像与查询图像高层语义特征之间的欧氏距离进行精检索，达到最终的检索目的。提出方法将哈希层的损失作为优化目标之一，结合图像的两种特征进行检索，弥补了现有方法中直接利用CNN深层特征检索耗时、占用内存的不足。在印花织物和CIFAR-10数据集上的实验结果表明，提出方法检索性能优于其他现有方法。

关键词: 图像检索, 卷积神经网络, 哈希编码, 分级检索