AbsoluteDeNoiser (presentation)


Latest news : NDNoise is a completely new write of AbsoluteDeNoiser from scratch

Description

AbsoluteDeNoiser is an easy to use software (step 1,2,3) that produces very competitive noise reduction filtering for digital images. It is a Java software that should run on almost every machines : Windows 98/2000/XP, Mac Os X, Unix, Linux, ...

Principle

Digital cameras or scanners often produce images that contains noise, especially when using high ISO. This noise is made of random pixel variations that damage the picture in different manner, adding unwanted granular effects. To obtain a good result, this noise should be removed paying a great attention to :
  1. Image details that often can be removed with noise, and edges definition that are often blured by basic noise reduction technics.
  2. Keeping some granularity to avoid the common "plastic" effect of too much noise removal.

AbsoluteDeNoiser (Free version) works in 3 steps :
  1. Edges detection : this analyse makes a separation between large surfaces with poor details, and places where some edges and/or details require a special attention. The value to tune enables to decide the sensitivity of the edges detection, thus the amount of the picture that is to be considered as details/edges or not. Surfaces and edges will then be processed in different ways in next steps.
  2. Surfacing : performs a calculation of local color cohesion to decide, for each pixel, what is the mean color value in the noise variations arround it, without merging two different surfaces together. This step also processes surfaces and edges in two different ways. The more important is the value that tunes the surface managment : it is the one that decides of the effective noise reduction. There is also a secondary value that tunes the edges managment, but it should often be kept at a low value to avoid bluring edges : in some cases it offers the possibility to reduce some edges pixelisation effects.
  3. Texturing : the step 2 may provide with "too" clean surfaces where some small details or textures could have been lost, and possibly a "plastic" effect. This final filter decides the amount of original pixels that should be added back to the clean surface to restore details and textures without re-introducing the noise that should be removed. A first value tunes how much part of pixels that was removed in step 2 shows a local regularity, thus how much noise vs details is to be considered on them. There is also a secondary value that tunes how much pure granularity, without special regularity, should be added back without re-introducing bad strong spots and scratches.

Examples and competitors

Michael Almond gives a noise reduction tool comparison page where this large list of tools (most of them are PhotoShop plugins, but there are also stand alone softwares) are tested :
His test images are low-noise digital photographs, but this is a good start point to show the kind of results AbsoluteDeNoiser can give. Here are my own results (with my own tuning.. with step 1,2,3.. others could have choosen different values) :

Strong noise removal :

Original
Result
cost_full.jpg
(1.6M)
Part (zoom 2x) :
Part
cost_full_ADN.jpg
(1.6M)
Part (zoom 2x) :
Part

mountain_glow_full.jpg
(1.9M)
Part (zoom 2x) :
Part
mountain_glow_full_ADN.jpg
(1.8M)
Part (zoom 2x) :
Part

valley_sunset_full.jpg
(1.6M)
Part (zoom 2x) :
Part
valley_sunset_full_ADN.jpg
(1.3M)
Part (zoom 2x) :
Part


Lower noise reduction keeping a little granularity :

Original
Result
cost_full.jpg
(1.6M)
Part (zoom 2x) :
Part
cost_full_ADNLow.jpg
(1.6M)
Part (zoom 2x) :
Part

mountain_glow_full.jpg
(1.9M)
Part (zoom 2x) :
Part
mountain_glow_full_ADNLow.jpg
(1.8M)
Part (zoom 2x) :
Part

valley_sunset_full.jpg
(1.6M)
Part (zoom 2x) :
Part
valley_sunset_full_ADNLow.jpg
(1.3M)
Part (zoom 2x) :
Part



Other competitors (new ones will be added in time) :
CleanerZoomer (new)
PictureCooler
NoiseWare
Helicon Noise Filter
Banding noise reduction for canon S40/G2 RAW files
DigiPhotoCleaner
Digital GEM Plug-in
ClearImage Repair
TurboPhoto
....??

Home