Monagi Alkinani, Winter 2017,
"Patch-based Denoising Algorithms for Single
and Multi-view Images"
Computer Science Department,
University of Western Ontario, Canada.
Ph.D. Thesis Abstract
In general, all single and multi-view digital images are captured using
sensors, where they are often contaminated with noise, which is an undesired
random signal. Such noise can also be produced during transmission or by lossy
image compression. Reducing the noise and enhancing those images is among the
fundamental digital image processing tasks. Improving the performance of image
denoising methods, would greatly contribute to single or multi-view image
processing techniques, e.g. segmentation, computing disparity maps, etc.
Patch-based denoising methods have recently emerged as the state-of-the-art
denoising approaches for various additive noise levels. This thesis proposes
two patch-based denoising methods for single and multi-view images,
A modification to the block matching 3D algorithm is proposed for
single image denoising. An adaptive collaborative thresholding filter is
proposed which consists of a classification map and a set of various
thresholding levels and operators. These are exploited when the collaborative
hard-thresholding step is applied. Moreover, the collaborative Wiener filtering
is improved by assigning greater weight when dealing with similar patches.
For the denoising of multi-view images, this thesis proposes algorithms that
takes a pair of noisy images captured from two different directions at the
same time (stereoscopic images). The structural, maximum difference or the
singular value decomposition-based similarity metrics is utilized for
identifying locations of similar search windows in the input images.
The non-local means algorithm is adapted for filtering these noisy
The performance of both
methods have been evaluated both quantitatively and qualitatively through a
number of experiments using the peak signal-to-noise ratio and the mean
structural similarity measure. Experimental results show that the proposed
algorithm for single image denoising outperforms the original block matching 3D
algorithm at various noise levels. Moreover, the proposed algorithm for
multi-view image denoising can effectively reduce noise and assist to estimate
more accurate disparity maps at various noise levels.