Denoising is the process of removing noise from a signal. Noise reduction techniques are conceptually very similar regardless of the signal being processed, however prior knowledge of the characteristics of an expected signal can mean the implementations of these techniques vary greatly depending on the type of signal.

In many applications, the acquired image is degraded by measurement or transmission noise. The noise can have different statistics, i.e., it can be additive noise, impulse noise, or block- or pixelwise losses.

Additive Gaussian Noise

Below is an example of an image degraded with additive Gaussian noise. On the left is the original image. In the middle is the degraded image, which has been contaminated with Additive White Gaussian Noise with a variance of 400. On the right is the result of image denoising using techniques proposed by the SIP group. The PSNR is 35.08.

For more information, see:
S. Voloshynovskiy, O. Koval, and T. Pun, "Image denoising based on the edge-process model", Signal Processing, vol. 85, iss. 10, pp. 1950-1969, 2005. [pdf|bib]

Uniform Noise

In the next example, we have a look at uniform noise. Again, the original image is on the left. Like the previous example, this noise is pixelwise, but, unlike the previous example, it is not additive. Instead, for 50% of the pixels have been replaced by completely random values. The denoised version is again on the right.

Salt-and-Pepper Noise

Defective sensors can cause so-called salt-and-pepper noise, meaning that some pixels are replaced by either completely white, or completely black pixels. Again the original image is on the left, a heavily degraded image where 50% of the pixels have been replaced by salt-and-pepper noise is in the middle, and the denoised version using techniques proposed by the SIP group is on the right.