Group- sparse signal denoising: Non- convex Regularization, Convex Optimization

Abstract

Standard way for assessing SNR is to use raised advancement with sparsity-advancing curved regularisation (signal to noise ratio). Non-raised enhancement is another common method in order to promote sparseness more clearly than arched regularisation. By using non-raised regularisation terms, the total cost of the task (which includes both information consistency and regularisation costs) is curved rather than flat. A more solid emphasis is placed on the concept of sparsity in this model, yet it retains the attractive aspects of arched augmentation (one of a kind least, vigorous calculations, and so forth.). For the denoising of small signals, we use this strategy to improve our GSS (group sub optimal shrinkage) computation. Both SNR and perceptual quality benefit from the calculation, which relates to the goal of improving dialogue.