------------------------------------------------------
Adaptive Noise Removal Filter, Version 1.1, README.TXT
------------------------------------------------------

Contents

System requirements
Description
New in version 1.1
How does it work?
Source code
Settings
Contact


-------------------
System requirements
-------------------

- Windows 95 or newer
- a fast CPU recommended

-----------
Description
-----------

The filter reduces noise in the image without blurring the edges. The
results are much better than can be achieved by using the popular adaptive
median or smoothing filters.Like modern compression algorithms (including JPEG) the filter operates in
Y,Cr, Cb color space, hence it is useful for removing compression artifacts.

------------------
New in version 1.1
------------------

- Added saving to JPEG
- Added adaptive median filtering for comparisons
- Released source code

-----------------
How does it work?
-----------------

At first the image is transformed from RGB to Y,Cr,Cb color space, then
a pixel neighbor graph is created with edge weights corresponding to 
color differences between the pixels. This graph is used to create a different 
filter for each pixel in each color plane.

- for the pixel being filtered calculate the shortest path from it to
all the other pixels reachable with path length < d
- replace the pixel value with the weighted average of all reached
pixels, weights are taken as the Gaussian function of the path length


This way only the relevant filtering neighborhood for each pixel is
selected. The size and shape of the neighborhood adapt to the image
features, including curved ones and sharp corners. If there is an edge in
the image, the proposed algorithm creates a filter that does not include
pixels in the opposite side of the edge. In this way edges are not smoothed
out.

Note that we select different maximum path length for each color plane 
resulting in different noise reduction level for each. 

-----------
Source code
-----------

The full source code is available under the terms of GNU General Public
License. Please see the homepage of the filter.
The program is compiled under Borland C++ Builder 5.0. The adoption to
other C++ compiler should not be hard.

--------
Settings
--------

You can adjust several parameters to obtain the best noise reduction level
for each particular image:

- Filter radius for each luma, Cr, Cb color channels: Larger radius will
remove more noise, however some detail also can be lost.
- Edge sensitivity: larger values will preserve more edges but sacrifice
smoothness in low contrast areas. For very large value banding may appear.
- Chroma 2x reduction: The resolution of Cr and Cb channels is reduced two
times, what improves the speed of the algorithm for large radiuses with
little loss of quality.

Recommended settings are.
- for high quality scanned images
radius luma 2, Cr 2 Cb 2
Edge Sensitivity 76
Chroma 2x reduction false

- for high quality JPEG images
radius luma 2, Cr 3 Cb 4
Edge Sensitivity 76
Chroma 2x reduction false

- for low quality JPEG images
radius luma 3, Cr 4 Cb 6
Edge Sensitivity 76
Chroma 2x reduction true

-------
Contact
-------

Please visit the website of the filter at:

http://www.gradetools.com/karlisf/image/filter.html

The author is waiting for your comments and questions by email:

Karlis.Freivalds@mii.lu.lv