Neat Image - Guide de l'Utilisateur / Profils de Bruit de Dispositif / Building
profile for specific device mode / Using
a regular image / Analyzing
image noise / Manual profiling
Advanced techniques available with manual profiling
As advised in the Étape
1. Preparing a regular image for noise analysis, the image used for noise
analysis should contain enough uniform featureless areas. If the input image
does not contain such areas then an alternative image from the same series can
be used. With manual profiling, there are several advanced profiling techniques
available that use one or two images. Below is the list of these techniques
sorted from the most desirable to the least desirable (from the standpoint of
profiling accuracy):
- Use a 100x100+ uniform featureless area in the input image for rough analysis;
then fine-tune the profile using several other uniform featureless areas in
the same image;
- Do (1) with an alternative image (may be an image of the Calibration Target
or a regular image) from the same camera (scanner) shot (scanned) in similar
conditions; then additionally fine-tune the profile using the input image;*
- Do (1) with another image from another camera (scanner) of the same model
shot (scanned) in similar conditions; then additionally fine-tune the profile
using the input image;
- Get a ready-made profile built with a similar image from another camera
(scanner) of the same model; then additionally fine-tune the profile using
the input image;
- Do (1) starting with a smaller (60x60-100x100) uniform featureless area
in the input image;
- Cut out a 59x59 or smaller uniform featureless area from the input image
and (preferably seamlessly, as much as possible) clone it in an image editor
to produce a 60x60+ area; do (1) with the resulting larger area;
- Up-sample your image (using your favorite resampling method) in an image
editor; do (1) with the up-sampled image; process the up-sampled image in
Neat Image (do not process original image with such a profile); down-sample
the result in the image editor.
* Here and below, when additional fine-tuning is applied using the input image,
the equalizer values should not change much if the two images were indeed shot
(scanned) in similar device modes. If the equalizer values do change a lot,
consider doing (1) with another image that is closer to the input image.
Copyright © Alex64 - 1998-2011 http://alex64.free.fr