ABSTRACT
Model selection in signal processing plays a crucial role to achieve
the underlying objective. Use of arbitrary or generalized model
without exploiting data prior results in poor performance. In
recent times, non-parametric methods such as dictionary-based
approach have been proposed for applications such as image
de-noising, in-painting, de-mosaicking, and compression. In this
chapter, we discuss regularized sparsity prior-based dictionary
learning algorithm named Regularized K Times Sum of Optimally
Weighted Vectors. We have mathematically formulated and derived
atom update expression for different priors. Dictionary learning
algorithm that considers a smoothing regularizer on dictionary
atom has been discussed for image de-noising. It has the advantage
of having closed form expression for atom update. The regularized
dictionary learning algorithm has been shown to achieve overall
better performance compared to some of the existing dictionary
learning techniques for image de-noising.
