ABSTRACT
However, most recent methods for dictionary learning are iterative batch processing algorithms, which deal with all the training samples at each iteration to minimize the objective function under sparse constraints. Therefore, another problem we may encounter is that when the training set becomes very large, these batch methods are no longer efficient. To overcome this bottleneck, an online algorithm for dictionary learning which applies stochastic approximation method has been proposed in the literature (Aharon et al. 2008, Mairal et al. 2010). To address these issues, we propose an incremental learning approach that processes one sample (or a small subset) of the training set at a time. This is particularly important in the context of image and video processing tasks, where it is common to learn dictionaries adapted to small image patches. The training samples set may include several millions of these patches. Our proposed approach is expected to efficiently handle large training set and effectively perform image denoising task.
