基于半监督学习的增量图像分类方法  

Incremental Image Classification Method Based on Semi-Supervised Learning

作　　者：梁鹏[1,2] 黎绍发[2] 覃姜维[2] 罗剑高[3] 

机构地区：[1]广东技术师范学院计算机科学学院,广州510665 [2]华南理工大学计算机科学与工程学院,广州510006 [3]广东农工商职业技术学院计算机系,广州510507

出　　处：《模式识别与人工智能》2012年第1期111-117,共7页Pattern Recognition and Artificial Intelligence

基　　金：国家自然科学基金项目(No.60273064);广东省工业攻关计划项目(No.2004B10101032);广东高校优秀青年创新人才培育项目(No.LYM09097)资助

摘　　要：为有效使用大量未标注的图像进行分类,提出一种基于半监督学习的图像分类方法.通过共同的隐含话题桥接少量已标注的图像和大量未标注的图像,利用已标注图像的Must-link约束和Cannot-link约束提高未标注图像分类的精度.实验结果表明,该方法有效提高Caltech-101数据集和7类图像集约10%的分类精度.此外,针对目前绝大部分半监督图像分类方法不具备增量学习能力这一缺点,提出该方法的增量学习模型.实验结果表明,增量学习模型相比无增量学习模型提高近90%的计算效率.In order to use large numbers of unlabeled images effectively,an image classification method is proposed based on semi-supervised learning.The proposed method bridges a large amount of unlabeled images and limited numbers of labeled images by exploiting the common topics.The classification accuracy is improved by using the must-link constraint and cannot-link constraint of labeled images.The experimental results on Caltech-101 and 7-classes image dataset demonstrate that the classification accuracy improves about 10% by the proposed method.Furthermore,due to the present semi-supervised image classification methods lacking of incremental learning ability,an incremental implementation of our method is proposed.Comparing with non-incremental learning model in literature,the incremental learning method improves the computation efficiency of nearly 90%.

关 键 词：半监督学习 图像分类 增量学习 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]