Cross-Domain Dictionary Learning For Image Classification

We address the image classification problem and present a method that utilizes labeled data from other image domains as the auxiliary source data for enhancing the original learning system.The proposed method aims to expand the intra-class diversity of original training data through the collaboration with the source data.In other words,we use the transfer learning method to share or transfer the information between the different domains or tasks.It can make traditional machine learning more flexible to learn.However,if the images which getting from the source domain as a supplement can not eliminate the difference between with the images that getting from the target domain after a conversion,at this point,we should consider the difference between the source domain and the target domain,so that we can provide better service to our target domain' classification.Dictionary learning has been successful in many computer vision tasks.In order to bring the original target domain data and the auxiliary source domain data into the same feature space,we introduce a cross-domain dictionary learning method,which learns reconstructive,discriminative and domain-adaptive dictionaries and the corresponding classifier parameters.Such a method operates at a high level,and it can be applied to different cross-domain applications.The Classification results of Transfer learning rely on the similarity of distribution between the source domains with target domain.Multisource domains transfer learning try to find the suitable samples from every different source domains to do the knowledge transfer.In this paper,we study the cross-domain dictionary learning for image classification.This paper focuses on solving the problem that the images which getting from the source domain as a supplement can't eliminate the difference between with the images that getting from the target domain after a conversion.Moreover,we use the multi-source thinking to further improve the accuracy of image classification.All works are done as follow:(1)We propose the cross-domain dictionary learning method for image classification with inter domain difference.Considering the difference between the source domain and target domain,retaining the difference between the source domain and the target domain.In reality,we obtain the source domain which is different from the target domain even after the migration and conversion.In this way,our method can better adapt to the reality,so that our image classification method is more widely used.(2)We propose the multi-source cross-domain dictionary learning method for image classification.The single-source is extended to multi-source.The application of multiple source domain data makes more target domain training data can be matched with the data in the source domain.In the training of dictionary and classifier,we can get multiple source domain dictionaries.So we can get the target domain dictionary and the target domain of the sparse coding better,so that better for image classification results.At the same time,we adjust each source domain relative proportion on the basis of experiment.It enables the source domain data to better enhance the image classification effect.