基于深度学习的图像推荐系统
首发时间:2017-05-31
摘要:如何提取具有代表性的图像特征并将其应用于图像检索,成为当前的研究热点之一。通过对图像信息的学习,卷积神经网络能有效提取丰富的图像特征,显著提高了各种图像处理任务的性能。本文设计并实现了基于深度学习的图像推荐系统。首先,通过深度学习框架Caffe提取图像的高维特征向量;然后运用局部敏感哈希算法思想在深度学习网络中将高维特征向量压缩为二进制哈希码,在MySQL数据库中存储高维特征向量和二进制码,并建立二进制码的索引;最后,系统在CIFAR-10数据集上进行了测试,实验结果表明本系统要优于基于KSH、CNNH、CNNH+、LSH等哈希算法的推荐系统。
关键词: 软件工程 深度学习 图像推荐 卷积神经网络 哈希算法
For information in English, please click here
Image recommendation system based on deep learning
Abstract:How to extract representative image features and apply them to image retrieval has become one of the focuses in the current research. Through the study of image information, convolutional neural network can effectively extract rich image features, and significantly improve the performance of various image processing tasks. This paper designs and implements the image recommendation system based on deep learning. Firstly, extracte the high-dimensional feature vectors of images by deep learning framework Caffe; Then compress the feature vectors into binary hash codes in the deep learning network by using the local sensitive hash algorithm thought, store the high-dimensional feature vectors and binary codes in the MySQL database, and build the index of binary codes; Finally, the system is tested on the CIFAR-10 dataset, and the results show that it is better than the recommendation systems based on KSH, CNNH, CNNH+, LSH and other hash algorithms.
Keywords: Software Engineering Deep Learning Image Recommendation Convolutional Neural Network Hash Algorithm
基金:
论文图表:
引用
No.4735036120230514****
同行评议
勘误表
基于深度学习的图像推荐系统
评论
全部评论0/1000