基于迁移学习的多标签图像标注  

Multi-label Image Annotation Based on Transfer Learning

作　　者：秦莹华 李菲菲[1] 陈虬 QIN Yinghua;LI Feifei;CHEN Qiu(Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《电子科技》2018年第8期21-24,共4页Electronic Science and Technology

基　　金：上海市高校特聘教授(东方学者)岗位计划(ES2012XX;ES2014XX)

摘　　要：自动图像标注(AIA)是图像检索领域一项具有挑战性的任务。卷积神经网络(CNNs)在大规模视觉识别挑战中取得杰出的分类性能。文中将图像标注问题看成一个多标签学习问题,提出了一种基于卷积神经网络的多标签分类算法。采用Corel5k数据库训练和测试卷积神经网络。由于训练集的缺乏,采用从Image Net分类任务中迁移得到的参数作为一个"中级图像特征的通用提取器"。采用两种损失函数作为多标签分类器,将多标签学习转化为多个标签的分类问题。将实验结果与其他方法进行比较,表明了该方法的有效性。Automatic image annotation（ AIA） is a challenging task in image retrieval field. Convolutional neural networks（ CNNs） have recently shown outstanding image classification performance in the large-scale visual recognition challenge（ ILSVRC2012）. The image annotation problem is considered as a multi-label learning problem and a multi-label classification algorithm based on convolutional neural networks is proposed. The Corel5 K image dataset is used to train and test the model. Due to the lack of training data,the parameters transferred from the Image Net classification task is used as a generic extractor of mid-level image representation. Two loss functions are used as multi-label classifiers and the multi-label learning problem are transformed into multiple labels classification problem. The experimental results are compared with other methods,and the effectiveness of the proposed method is demonstrated.

关 键 词：自动图像标注 多标签 卷积神经网络 迁移学习 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]