FP-CNNH:一种基于深度卷积神经网络的快速图像哈希算法

FP-CNNH:A Fast Image Hashing Algorithm Based on Deep Convolutional Neural Network

在大数据时代,图像检索技术在大规模数据上的应用是一个热门的研究领域。近年来,大规模图像检索系统中,图像哈希算法由于具备提高图像的检索效率同时减少储存空间的优点而受到广泛的关注。现有的有监督学习哈希算法存在一些问题,主流的有监督的哈希算法需要通过图像特征提取器获取人为构造的图像特征表示,这种做法带来的图像特征损失影响了哈希算法的效果,也不能较好地处理图像数据集中语义的相似性问题。随着深度学习在大规模数据上研究的兴起,一些相关研究尝试通过深度神经网络进行有监督的哈希函数学习,提升了哈希函数的效果,但这类方法需要针对数据集人为设计复杂的深度神经网络,增大了哈希函数设计的难度,而且深度神经网络的训练需要较多的数据和较长的时间,这些问题影响了基于深度学习的哈希算法在大规模数据集上的应用。针对这些问题,提出了一种基于深度卷积神经网络的快速图像哈希算法,该算法通过设计优化问题的求解方法以及使用预训练的大规模深度神经网络,提高了哈希算法的效果,同时明显地缩短了复杂神经网络的训练时间。根据在不同图像数据集上的实验结果分析可知,与现有的基准算法相比,提出的算法在哈希函数训练效果和训练时间上都具有较大的提高。

In the big data era, application research on image retrieval technology for large-scale data is a hot field. Recent years, image hash algorithm has attracted much attention in large-scale image retrieval system in order to improve the retrieval efficiency and reduce the storage space. However, there are some issues with existing supervised hash code learning algorithms. Most of the supervised hash algorithms need to use image feature extractor for obtaining hand-craf- ted image features, which influence the effect of hash code training with image features, at the same time these methods cannot deal well with semantic similarity for image data set. With the development of deep learning research on the large-scale data, some recent related work try to deploy deep neural network to learn hash function and make the hash code training effect increased. But such kind of methods require carefully designed complex neural network structure thus increase the difficulty of the hash function design and cost more time on neural network training with large data set. These problems limit the range of the hash algorithm application with the deep learning architecture for large data sets. To solve the problems mentioned above, this paper proposed a fast image hashing algorithm based on deep convolu- tion neural network. The proposed algorithm consists of an optimization approach for constructing the hash code of the training data set and a pre-trained large deep neural network for learning to improve the effect of hash algorithm, shor- tening the training time of complex neural network. According to the analysis of experimental results on different image data sets, both the effectiveness of hash function and the efficiency of training time of the proposed algorithm have bet- ter oerformance corot）areal with the existing algorithms.