Design of Efficient Clustering Dictionary for Sparse Representation of General Image Super Resolution

Designing an efficient over-complete dictionary is an important issue for developing a learning based system of super-resolution. To obtain fast solution, the size of dictionary needs to be reduced. However it may lower the performance as dictionary maybe incomplete. To address this issue, in this paper, we propose an improvement of dictionary learning for image super-resolution based on sparse representation. The proposed training process method can generate an efficient clustering of an over-complete dictionary without reducing its size but with low computational cost when applied with basis pursuit denosing of sparse representation solution. The performance of the proposed dictionary is satisfactory in terms of computational time reduction with comparable RMSE (root mean square error) when compared with other known methods.