A Sparse Dictionary Learning Algorithm with BB Method

In recent years, more dictionary learning algorithms have been proposed, and the Lagrange dual method achieves better performance. It is an optimization algorithm with the gradient descent methods, and therefore it is inevitable to deal with the step size problem. In this paper, we first introduced the Lagrange dual method thoroughly, and then the step size problem is discussed in detail. Next, a sparse coding dictionary updating algorithm with BB method is proposed: combined with the two-point step size method (or BB method) proposed by Barzilai and Borwein, the Lagrange dual method is applied to solve the problem of step size adjustment. The proposed algorithm is superior to other methods in many data sets as shown by the experimental results.