基于字典对齐的迁移稀疏编码图像分类

基于稀疏编码的图像分类算法，当源域和目标域间样本服从不同分布时，从源域样本中学习到的字典无法有效对目标域样本进行编码，进而严重影响算法的分类性能。为了解决此问题，提出一种基于字典对齐的迁移稀疏编码(TSC-DA)算法。一方面，通过将字典对齐机制引入稀疏编码模型训练过程中，以减少源域和目标域间样本分布差异；另一方面，采用L2正则化项代替字典约束项，将其转化为无约束优化问题，从而回避了拉格朗日对偶法复杂的求解方式。实验结果表明，TSC-DA能够有效提高目标域的图像分类精度。

Transfer sparse coding for image classification based on dictionary alignment

In traditional sparse coding classification model,when the samples between source and target domains are subject to different distributions,the dictionary learned from source domain samples can not effectively encode target domain samples so as to seriously degrade classification performance.To solve the problem,a Transfer Sparse Coding based on Dictionary Alignment(TSC-DA) algorithm is proposed.On the one hand,through introducing dictionary alignment method into the training process of sparse coding model,distribution of samples in different domains is reduced.On the other hand,in order to avoid complex lagrange dual method,L2 regularization term is used to replace original dictionary constraint term,and then it can transform into an unconstrained optimization problem.Experimental results show that TSC-DA can effectively improve the target domain image classification accuracy.