Ximagic Denoiser toolbox
Ximagic Denoiser is a Photoshop plugin for image noise reduction.
It provides seven different methods for noise reduction
- Median (Std/Center Weighted)
- Gauss (Std/Bilateral)
- DWT (Overcomplete Wavelets)
- CWT (Complex wavelets)
- DCT (Discrete cosine)
- NLM (Non Local Means)
- Non-linear (anisotropic) diffusion (Curvature & Gradient)
Support of 8/16 bits images.
Support of Photoshop actions/batch
Support use as Smart filter
Versions for 32 and 64 bit plugin hosts
Mac and Windows versions
Compatible with Paint Shop Pro, The Gimp, IrfanView, XnView, ...
In the area of image noise reduction, there are more than one valid algorithm, it depends on the type of the image (artistic photos, medical image, etc), the noise the image has (salt & pepper noise, shot noise, Gaussian, uniform, grain etc) and the desired result (precision, artistic, descriptive, etc)
- Local Blurring
Generally speaking, the first methods used was median and Gaussian blur, the first one is good for salt and pepper noise and the second is of more general use. Those first methods blur the image and the edges of it. Those methods can be called local blurring because it only take into account a small number of pixels near to the processed one.
Blurring the edges of a image is not usually accepted, and then come the bilateral filter which take into account the distance from each point to the processed one in color and position. This is probably the method more usual today.
- Decomposition filtering
Another kind of methods was based on decomposing the image and filtering it, FFT wavelets and DCT. From those methods, wavelets are actually more used. It gives a good relation between smoothing and preservation of edges.
- Non local Means
A recent method which not only reduces noise but also defects and scratches. For medium/low noise images this is the best method.
- Nonlinear Diffusion
This family of methods are based on the diffusion equation. Those methods are used mainly in the area on medical image processing. The included one is Anisotropic diffusion.
Usually if the image have medium/low noise, non local means give the best results.
For high noise images usually a combination of wavelets and non-local means give the best results.
Questions, suggestions and bug reports
Comments are welcomed. Please e-mail to firstname.lastname@example.org
Thanks, reused code & copyrights
This application reuses some published code, I wish to thanks the Authors or Publishers
- ITK Insight Toolkit [anisotropic diffusion]
- Onur G. Guleryuz [wavelets and DCT]
- Intel Threading Building Blocks
- TinyXML parser
To see the detail of code used and the copyrights go to License