From High Definition Image to Low Space Optimization

Signal and image processing have seen in the last few years an explosion of interest in a new form of signal/image characterization via the concept of sparsity with respect to a dictionary. An active field of research is dictionary learning: Given a large amount of example signals/images one would like to learn a dictionary with much fewer atoms than examples on one hand, and much more atoms than pixels on the other hand. The dictionary is constructed such that the examples are sparse on that dictionary i.e each image is a linear combination of small number of atoms. This paper suggests a new computational approach to the problem of dictionary learning. We show that smart non-uniform sampling, via the recently introduced method of coresets, achieves excellent results, with controlled deviation from the optimal dictionary. We represent dictionary learning for sparse representation of images as a geometric problem, and illustrate the coreset technique by using it together with the K−SVD method. Our simulations demonstrate gain factor of up to 60 in computational time with the same, and even better, performance. We also demonstrate our ability to perform computations on larger patches and high-definition images, where the traditional approach breaks down.