|
Title:
|
MACHINE LEARNING CONTENT ADAPTIVE FILTERS
FOR IMAGE DE-BLURRING |
|
Author(s):
|
Pradip Mainali |
|
ISBN:
|
978-989-8533-91-3 |
|
Editors:
|
Katherine Blashki and Yingcai Xiao |
|
Year:
|
2019 |
|
Edition:
|
Single |
|
Keywords:
|
De-blurring, Deconvolution, Restoration |
|
Type:
|
Full Paper |
|
First Page:
|
215 |
|
Last Page:
|
222 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
This paper proposes a novel algorithm to recover blindly a sharp image from its degraded form by pixel pattern
classification and filtering by using the deconvolution filters trained for the corresponding pixel pattern. Noise
amplification is a well-known phenomenon that arises due to an ill-posed nature of the deconvolution, which is tackled
better with the pattern classification, thereby state-of-the-art image de-blurring quality is demonstrated, yet at the very
low complexity. For the computational efficiency, the pixel pattern classes are learnt in a multi-layer structure consisting
of the coarser and the finer patterns, reducing the pattern search complexity in the dictionary by a factor of 6 without
losing any quality. As compared to deep-learning based methods, the proposed approach provides a smaller memory
footprint and lightweight implementation. For lightweight and real-time implementation, the paper also validates a simple
class matching metric of l1-norm. The proposed method is suitable for embedded applications such as camera, TV
systems, etc. |
|
|
|
|
|
|
