Abstract

This paper presents a generic optimization framework for efficient fea- ture quantization using sparse coding which can be applied to many computer vi- sion tasks. While there are many works working on sparse coding and dictionary learning, none of them has exploited the advantages of the marginal regression and the lasso simultaneously to provide more efficient and effective solutions. In our work, we provide such an approach with a theoretical support. Therefore, the computational complexity of the proposed method can be two orders faster than that of the lasso with sacri ficing the inevitable quantization error. On the other hand, the proposed method is more robust than the conventional marginal regres- sion based methods. We also provide an adaptive regularization parameter se- lection scheme and a dictionary learning method incorporated with the proposed sparsity estimation algorithm. Experimental results and detailed model analysis are presented to demonstrate the efficacy of our proposed methods.