White noise with a std. dev. of 35 will be added to the Mandrill image.
The denoising methods will attempt to remove the added noise.
The main challenges of the image will be the eyes and hair.
The hair provides a great example of a natural texture.
Mandrill image
Add Gaussian white noise with std. dev. =
35
Gaussian filter
The denoised image is blurry and still contains nosie. The hair has no
direction. The
method noise contains a large amount of detail and structure from
the image.
Denoised image
Method noise
MSE=383.2
Wiener filter
The denoised image has less noise, but it is still blurry. The hair
has direction, but no detail. The
method noise contains detail and structure from
the image.
Denoised image
Method noise
MSE=395.4
SUSAN filter
The denoised image is less blurry and noisy.
The hair has some direction and shows some detail.
The method noise contains detail and structure
from the image.
Denoised image
Method noise
MSE=343.2
Non-local means
The denoised image is less blurry and noisy, but it does have
some contouring at the bottom. The hair has flow and detail. The area
around the eyes contains little noise. The
method noise contains very little
structure from the image.
