Barb Test Case

White noise with a std. dev. of 25 will be added to the Barb image. The denoising methods will attempt to remove the added noise. The main challenges of this image will be the face and stripes on the clothes.

Barb image
Add Gaussian white noise with std. dev. = 25


Gaussian filter

The denoised image is blurry and still contains nosie. No stripes are on the pants. The method noise contains a large amount of detail and structure from the image.

Denoised image
Method noise
MSE=297.7


Wiener filter

The denoised image is still blurry and noisy. The stripes are handled better. The method noise contains a large amount of detail and structure from the image.

Denoised image
Method noise
MSE=186.7


SUSAN filter

The denoised image is less blurry and noisy. The face and stripes are clearer. The method noise contains detail and structure from the image.

Denoised image
Method noise
MSE=162.9


Non-local means

Compared to the other methods the denoised image has little blurring and noise, but it does have some contouring on the arm for example. The stripes are easily reproduced by the non-local means method because of their periodic nature. The method noise contains very little structure from the image.

Denoised image
Method noise
MSE=90.5