1: \begin{abstract}

2: 

3: A new method is proposed in this paper to learn overcomplete dictionary

4: from training data samples. Differing from the current methods that

5: enforce similar sparsity constraint on each of the input samples, the

6: proposed method attempts to impose global sparsity constraint on the

7: entire data set. This enables the proposed method to fittingly assign the

8: atoms of the dictionary to represent various samples and optimally adapt

9: to the complicated structures underlying the entire data set. By virtue of

10: the sparse coding and sparse PCA techniques, a simple algorithm is

11: designed for the implementation of the method. The efficiency and the

12: convergence of the proposed algorithm are also theoretically analyzed.

13: Based on the experimental results implemented on a series of signal and

14: image data sets, it is apparent that our method performs better than the

15: current dictionary learning methods in original dictionary recovering,

16: input data reconstructing, and salient data structure revealing.

17: \end{abstract}