0

I'm looking to remove unwanted halo artifacts from an image, that is located in the LAB colorspace.

So my Question is: Given an image I in Lab-Colorspace, the gauss blurred L-Channel I(blur), a sharped L-Channel I(sharp), and the resulting sharpened Original L-Channel I(end).

How can I distract Halo-Artifacts from my resulting sharpened Original I(end), that mustnt be there?

I(end)=I+y*(I-I(blur));

The original image

enter image description here

The halos appear after sharpening the image with a gaussian filter. I have been searching for days for a useful attempt. As somebody wrote here vignetting is NOT equal to Halo-Artifacts.

Therefore I'm looking to implement an algorithm that uses only the L(uminance)-Channel within my picture.

Heres what I get:

enter image description here

Heres the sharp-mask I(sharp)=I-I(blur). As u can see there are lightnings produced at the transition of the building and the sky. how can i manage to remove those artifacts? Can I filter them also anyhow? mayybe another gauss-filter? enter image description here

Heres my MATLAB code so far, resulting in halo artifacts.

imageOrig = imread('HinduTemple_Small.tif');
imdisplay(imageOrig, 'RGB  orginial',1);
[x,y,u]=size(imageOrig);
% Umwandeln in XYZ
simXYZ = imColorTransform (imageOrig,'ICCProfiles/AdobeRGB1998.icc',   
'*XYZ');

%Umwandeln von XYZ zu Lab
labOrig = imXYZ2Lab (simXYZ);
 labSharpened=labOrig; 

%Separieren des Luminance-Channels
lChannel=labOrig(:,:,1);

%Erzeugen eines Tiefpass-Filters, ich habe mich für einen Gaussfilter
%entschieden, da dieser eine zuverlässige Filterung im Gegensatz zum
%Average-Filter verspricht, siehe hierzu 
%https://de.wikipedia.org/wiki/Gleitender_Mittelwert#/media  
 /File:FrequenzSweepGlMWBinomial7.png

h = fspecial('gaussian', [100 100], 100);
lChannelFilter = imfilter(lChannel, h);
Gamma=0.5;
LChannelSharp=lChannel-lChannelFilter;
LChannelSharp= LChannelSharp;
lChannelnew=lChannel+Gamma*LChannelSharp;

labSharpened(:,:,1)=lChannelnew(:,:);
RGBSharp=imColorTransformMac(AntihalolabSharpened,'*Lab','ICCProfiles/AdobeRGB1998.icc');
 imdisplay(Sharp, 'RGBsharpened with HALO local Contrasted',1);
noob
  • 23
  • 7
  • What is your question? Please take a [tour](http://stackoverflow.com/tour). – Peter Mar 07 '16 at 13:50
  • Updated for better understanding: Given an image I in Lab-Colorspace, the gauss blurred L-Channel I(blur), a sharped L-Channel I(sharp), and the resulting sharpened Original L-Channel I(end). How can I distract Halo-Artifacts from my resulting sharpened Original I(end), that mustnt be there? – noob Mar 07 '16 at 14:04
  • MY guess is that what you get is as good as it gets. Either you find a research paper that deals with this, or you invent a new method, and do publish it of thats the case. – Ander Biguri Mar 07 '16 at 15:20
  • yes, im looking forward to implement the retinex-algorithm if i wont succeed in anyway...but to be honest im not sure if it works with Lab-colorspace. One thought is to look where the color differs from the original and to replace those pixels with those from the original image. But i cant figure out how to write the rule...because im only working on the luminance channel there shouldnt be colorchanges. But there are!? – noob Mar 07 '16 at 15:29
  • Nobody? I Have to return this tomorrow at the University "TH Cologne"... – noob Mar 08 '16 at 15:42

0 Answers0