Document Reading Notes -- group sparsity and geometry constrained dictionary

On Friday, some students reported a paper in iccv2013, group sparsity and geometry constrained dictionary learning for action recognition from depth maps. This article is about sparsing coding. Sparse Coding is not my research direction. Here is just a Document Reading Note, right to broaden my horizons.

From the title, we can see that this paper attempts to use the group sparsity and geometry constrained dictionaries for behavior recognition of deep graphs. Like the Sparse Coding paper we have seen, this article also uses a beautiful example to show how this algorithm differs from kmeans and general Sparse Coding algorithms. For example:

The two subgraphs (a) and (B) in the graph are more common than those in the plot. The subgraph (c) is Sparse Coding + group sparsity. From the graph, it is not difficult to see that group sparsity refers to the representation of all kinds of samples in only various subdictionaries. Subgraph (d) is an algorithm proposed by the author. Compared with (c), subgraph (d) adds geometric limitations, that is, the distance of the samples in the original feature space is close after sparse representation, that is, try to maintain the Geometric Distance between samples in the original space.

After understanding the author's intention, it is not difficult to understand the model created in this article. Compared with the previous work of researchers, this model adds the last one. Since the added item is convex, the addition of one item still maintains the convex Property of the function, so the model optimization method is the same as the general Sparse Coding Algorithm.

Document Reading Notes -- group sparsity and geometry constrained dictionary