Multi-morphology image super-resolution via sparse representation

• An image super-resolution algorithm based on morphological diversity is proposed.
• The learning mechanism relies on sparse representation.
• Dictionary training algorithm improves accuracy and efficiency for representation.
• An effectively sampling by variance is adopted to construct training set.
• Good performance can be achieved at low dictionary size and patch size.

A novel multi-morphology image super-resolution algorithm via sparse representation is proposed in this paper. Firstly, the observed low resolution image is decomposed into a linear combination of morphological components by morphological component analysis. Secondly, for effective representation of different components, each dictionary pair is learned from the low resolution component and corresponding high resolution component training sets. Then, each high resolution morphological component is generated by combine the sparse coefficients of low resolution morphological component with the corresponding high resolution component dictionary. Finally, fusing all high resolution morphological components yields the final high resolution image. Experimental results on various natural and intelligent transport systems surveillance images demonstrate the superiority of the proposed method in terms of qualitative and quantitative performance.