Comparing
Performance Measures of Sparse Representation on Image Restoration Algorithms
Subramaniam Sakthivel1, Parameswari Marimuthu2 and Natarajan Vinothaa3
1Department of Computer Science and Engineering, Sona College of Technology, India
2 Department of Computer Science and Engineering, Dhirajlal Gandhi College of Technology, India
3 Department of Computer Science and
Engineering, Sri Ramakrishna Engineering College, India
Abstract: Image restoration is a systematic process that
regains the lost clarity of an image. In the past, image restoration based on
sparse representation has resulted in better performance for natural images.
Within each category of image restoration such as de-blurring, de-noising and
super resolution, different algorithms are selected for evaluation and
comparison. It is evident that both local and non-local methods within each
algorithm can produce improved image restoration results based on the over
complete representations using learned dictionary. The Gaussian noise is added
with the original image and comparative study is made from the three different
de-noising techniques such as mean filter, Least Mean Square (LMS) adaptive
filters and median filters. The experimental results arrived from the filters
are discussed for each model of the selected image restoration algorithms-locally
adaptive sparsity and regularization, Centralized Sparse Representation (CSR), low-rank
approximation structured sparse representation and non-locally CSR. A
comprehensive study of this paper would serve as a good reference and stimulate
new research ideas in Image Restoration (IR).
Keywords: IR, sparse representation,
image de-blurring, locally adaptive sparsity, CSR.
Received
February 22, 2014; accepted July 9, 2014