基于深度学习的图像标签生成算法教程.docVIP

基于深度学习的图像标签生成算法 杨阳 张文生 （中国科学院自动化研究所北京 100190）摘要 图像的自动标注是图像检索领域一项基础而又富有挑战性的任务当前机器学习算法在该领域取得了一定的进展。深度学习算法自提出以来在图像和文本识别领域取得了巨大的成功，它可以作为一种解决“语义鸿沟”问题的有效方法。文中视为一个多学习问题图像的标签为信息最后Corel和ESP图像数据集验证了该模型框架及所提出解决方案的有效性。 关键字 图像自动标注，深度学习，神经网络，机器学中图分类号 TP39 Image Auto-Annotation based on deep learning Yang Yang, Zhang Wensheng (Institute of Automation, Chinese Academy of Sciences) Abstract Image auto-annotation is a basic and challenge task in the image retrieval work. The traditional machine learning methods have got a lot achievement in this field. The deep learning algorithm has achieved great success in image and text learning work since it has been presented, so it can be an efficiency method to sole the semantic gap problems. Image auto-annotation can be decomposed into two steps: basic image auto-annotation based on the relationship between image and tag, and annotation enhanced based on the mutual information of the tags. In this article, the basic image auto-annotation is viewed as a multi-labelled problem, so the prior knowledge of the tags can be used as the supervise information of the deep neural network. After the image tags got, the dependent relationship of the tags is used to improve the annotation result. At the end, this model have been tested in Corel and ESP dataset, and been proved that this method can efficiently solve the image auto-annotation problems. Key words image auto-annotation, deep learning, neural network, machine learning 1 引言 大数据时代人们可以访问获取的现爆炸式的增长，互联网上每天都诞生海量的图像和视频。为了有效地组织、查询与浏览如此大规模的图像资源图像检索技术应运而生。现有的图像检索方式主要分为两种基于内容的图像检索(Content-Based Image Retrieval, CBIR)和基于文本的图像检索(Text-Based Image Retrieval, TBIR)对TBIR来说它要求用户提交文本作为查询图像需要事先[1]。图像的ag标签一种有效获取图像的语义信息的方法，它应用广泛，。互联网上同存在大量的图像，人们虽然图像标签的自动生成是一个十分困难的任务，但是通过学者们的不断努力，取得了丰硕的成果。图像标签按照标注模型的不同主要分成两大类[2]基于统计分类的自动图像标注模型和基于概率的自动图像标注模型。基于统计分类的方法是将每一个图像的语义概念都被当作一个类别进行分类，自动图像标注就可以转换成图像的多分类问题。代表方法有:支向量机(Support Vector MachineSVM)方法[34，24]、二维多分辨率马尔可夫模型(2D MHMMs)[5]，贝叶斯点学习机(Bayes oint Machine)[6]和混合分级模型(Mixture Hi