Correcting Sample Selection Bias for Image Classification
One of the basic assumptions in traditional machine learning is that it requires training and test data be under the same distribution.However,in image classification,this assumption often does not hold,since image labels are not as sufficient as text ones.In this paper,we propose to use labeled images from relevant but different categories to take the role oftraining data for estimating a prediction model. Correcting sample selection bias,the 2000 Nobel Prizework in Economic,is applied to our problem.Weassume that the difference between training and testdata is that they are governed by different distributions.By eliminative sample selection bias in the trainingdata,the supervisory knowledge in the training datacan be effectively learned for classifying images in thetest set.We present theoretical and empirical analysisto demonstrate the effectiveness of our algorithm.Theexperimental results on two image corpora show thatour algorithm can greatly improve several state-of-the-art classifiers when the training and test images comefrom similar but different categories.
Di Wu Dinzhong Lin Li Yao Wenjun Zhang
Institute of Image Communication and Information Processing,Department of Electronic Engineering,Sha Department of Automation,Xiamen University,Xiamen 361005,China College of Software Engineering,Southeast University,Nanjing 211189,China
国际会议
厦门
英文
1214-1220
2008-11-17（万方平台首次上网日期，不代表论文的发表时间）