Performance analysis of unconventional dictionary on retinal images

Signals are sparsely represented by a set of over-complete basis vectors called dictionary. Atoms of a dictionary are either chosen from a predefined set of functions such as Discrete Cosine Transform (DCT), Wavelets, Gabor, Contourlets, and Curvelets or learned from an image itself. Recently, a nonlinear (NL) dictionary has been proposed by adding NL functions to the conventional DCT atoms. This paper presents performance of a novel NL dictionary for retinal image reconstruction and denoising. The NL dictionary demonstrates good performance and can be adapted if shorter execution time is required.