LAUSR

Conventional representation based classification methods, such as sparse representation based classification (SRC) and collaborative representation based classification (CRC) have been developed and shown great potential due to its effectiveness in various recognition tasks. However, when the test data and training data come from different distribution, the performance of SRC and CRC will be degraded significantly. Recently, several sparse representation based domain adaptation learning (DAL) methods have been proposed and achieve impressive performance. However, these sparse representation based DAL methods need to solve the ℓ1-norm optimization problem, which is extremely time-consuming. To address this problem, in this paper, we propose a simple yet much more efficient domain adaptive collaborative representation-based classification method (DACRC). By replacing the ℓ2-norm regularization term using the ℓ2-norm, we exploit the collaborative representation rather than sparse representation to jointly learn projections of data in the two domains. In addition, a common dictionary is also learned such that in the projected space the learned dictionary can optimal represent both training and test data. Furthermore, the proposed method is effective to deal with multiple domains problem and is easy to kernelized. Compared with other sparse representation based DAL methods, DACRC is computationally efficient and its performance is better or comparable to many state-of-the-art methods.