I have taken the liberty of running the page through an automatic translation page. Rough translation of http://observatory.playland.ru/article21_article_23_1.phtml
Please be aware of possible translation errors, especially in the bit about distribution!!
[edit] added stuff from the included readme.
==============================================================
EnhancePAL v.0.1: the filter for Avisynth 2.5
Reduces the characteristic distortion for captures made in the PAL standard which shows as a shift in the hue of two successive lines - one towards a red hue, another towards green.
Download:
download (http://observatory.playland.ru/soft/EnhancePALMMXSSE2.rar) the version optimized under MMX and SSE2.
download (http://observatory.playland.ru/soft/EnhancePAL%20sources.rar) the sourcecode of the filter.
Usage:
This version of the filter only supports YUY2.
In a script it is necessary to register a plug-in:
LoadPlugin ("<plug-in folder>\EnhancePAL.dll ")
Syntax:
EnhancePAL ([mode] (0), (0.5), [v_weight] (0.5), [chroma_thresh] (255))
mode - operating mode of the filter. Can accept values 0, 1, 2. Default is 0.
0 - averages the chroma between two rows and moves on to the next pair. Thus, distortion is completely corrected, but decreases color vertical detail by two. This mode is similar to algorithm of the filter rmPAL for VirtualDub,
(http://homepages.fh-giessen.de/~hg6423/rmPal/)
1 - for each pixel the chroma value is averaged with the chroma component from the pixel located under it using a weight factor. The color components contribute as follows:
U_new = U_old * (1-u_weight) + U_bottom * u_weight
V_new = V_old * (1-v_weight) + V_bottom * v_weight
Obviously the bottom line remains untouched.
2 - a mode to check an idea by Eugen65 concerning the calculation of correction values for a component. It simply calculates an average chroma difference for the croma components of two vertical points in consecutive lines. If the difference is more than the value chroma_thresh the difference is considered to be caused by noise, and the result is not used.
If value chroma_thresh is set to 255, all values are used.
u_weight - Weight factor for a mode 1. Default value is 0.5
The weight factors are used as in mode 0, but in this mode any factor other than 0.5 has no physical sense.
v_weight – Also a weight factor, but for V components. Default value is 0.5.
chroma_thresh - the threshold for the calculation of average deviations in a mode 2. Value by default 255. See an explanation of a mode 2
[u]Example:
EnhancePAL ()
* a mode 0. Works as rmPAL (the truth, is much faster)
EnhancePAL (1, 0.3, 0.3)
* a mode 1. Weight factors - on 0.3
Known Problems:
At work with several filters in a script (AviSynth 2.5.3) at moving
On a clip a toddler more than on one staff the filter crashes. But it is stable in ависинте,
I know as it to bypass, but have already moved to AviSynth 2.5.4, in which this does not happen.
Aknowledgements:
Thanks to Eugeny Suchevu aka Eugen65 for an explanation of the theory of the cause of the given distortion and discussions.
Distribution:
The program is distributed freely and as-is under GNU-GPL. An additional request - if anyone changes something, please inform the author to keep the program in a synchronous condition so it will be more usefull for all who possibly want to use the filter.
Also the request to inform on the noticed bugs, opiskah, etc.
The author of the filter - Ilya Utkin aka Duck
Optimization under MMX and SSE2 - Victor Tomilov aka VicoNT
Page of the filter: http://observatory.playland.ru/article21_article_23_1.phtml
Discussion of the Filter: http://forum.playland.ru/topic1754.phtml