Are there any limitations of CLAHE? (contrast limited adaptive histogram equalization)

[lisawa...@gmail.com]

Are there any limitations of CLAHE, contrast limited adaptive histogram equalization?

CLAHE significantly improves local contrast of images. I wonder if it has any limitation or downsides? Does it introduce some artifacts for some images? Are there any good examples?

Thanks,

Lisa

[aruz...@general-cathexis.com]

On Thursday, September 20, 2012 6:43:19 PM UTC-5, lisawa...@gmail.com wrote:
> Are there any limitations of CLAHE, contrast limited adaptive histogram equalization? CLAHE significantly improves local contrast of images. I wonder if it has any limitation or downsides? Does it introduce some artifacts for some images? Are there any good examples? Thanks, Lisa

According to

http://www.ijcse.com/docs/INDJCSE12-03-01-071.pdf

the disadvantages are "Noise amplification in flat region and
ring artifacts at strong edges."

Test it yourself with ImageJ and a CLAHE plugin available at

http://rsb.info.nih.gov/ij/plugins/clahe/index.html

[aruz...@general-cathexis.com]

On Thursday, September 20, 2012 6:43:19 PM UTC-5, lisawa...@gmail.com wrote:

> Are there any limitations of CLAHE, contrast limited adaptive histogram equalization? CLAHE significantly improves local contrast of images. I wonder if it has any limitation or downsides? Does it introduce some artifacts for some images? Are there any good examples? Thanks, Lisa

Authors in image processing jourmnals often select test images to make their algorithms look better than they are. Here is a test image that I selected to bring out artifacts, other than noise, in algorithms for increasing contrast,

http://www.general-cathexis.com/Manual/CarKabob.png

Here is the result from applying the CLAHE JImage plugin with default settings:

http://www.general-cathexis.com/images/CarKabobCLAHE.png

Here is a result of my proprietary "hysteresis" method:

http://www.general-cathexis.com/Manual/HysteresisContrastEnhanced6.jpg

Here is a result of the Farbman method:

http://www.general-cathexis.com/Manual/FarbmanContrastEnhanced.jpg

See http://www.cs.huji.ac.il/~danix/epd/ .

In my opinion, CLAHE is crap.

[Martin Leese]

aruz...@general-cathexis.com wrote:
> On Thursday, September 20, 2012 6:43:19 PM UTC-5, lisawa...@gmail.com wrote:
>> Are there any limitations of CLAHE, contrast limited adaptive histogram equalization? CLAHE significantly improves local contrast of images. I wonder if it has any limitation or downsides? Does it introduce some artifacts for some images? Are there any good examples? Thanks, Lisa

I have never used CLAHE, but in general:

All non-trivial algorithms introduce
artifacts. The trick is to choose an
algorithm such that the artifacts which
are introduced do not interfere with the
information you wish to enhance.

> Authors in image processing jourmnals often select test images to make their algorithms look better than they are. Here is a test image that I selected to bring out artifacts, other than noise, in algorithms for increasing contrast,
>
> http://www.general-cathexis.com/Manual/CarKabob.png
>
> Here is the result from applying the CLAHE JImage plugin with default settings:
>
> http://www.general-cathexis.com/images/CarKabobCLAHE.png

Wonderful artifacts; very clear.

> Here is a result of my proprietary "hysteresis" method:
>
> http://www.general-cathexis.com/Manual/HysteresisContrastEnhanced6.jpg

More subtle artifacts, particularly the
difference in the sky either side of the
left-hand (cellphone) pole. Also, the
saturation in the image appears to increase;
this may or may not be appropriate.

> Here is a result of the Farbman method:
>
> http://www.general-cathexis.com/Manual/FarbmanContrastEnhanced.jpg

Uniform sky; little change to saturation.

> See http://www.cs.huji.ac.il/~danix/epd/ .
>
> In my opinion, CLAHE is crap.

On this image it certainly is. On other
images it may be appropriate.

--
Regards,
Martin Leese
E-mail: ple...@see.Web.for.e-mail.INVALID
Web: http://members.tripod.com/martin_leese/

[aruz...@general-cathexis.com]

On Tuesday, September 25, 2012 10:37:01 AM UTC-5, Martin Leese wrote:
> aruz...@general-cathexis.com wrote: > On Thursday, September 20, 2012 6:43:19 PM UTC-5, lisawa...@gmail.com wrote: >> Are there any limitations of CLAHE, contrast limited adaptive histogram equalization? CLAHE significantly improves local contrast of images. I wonder if it has any limitation or downsides? Does it introduce some artifacts for some images? Are there any good examples? Thanks, Lisa I have never used CLAHE, but in general: All non-trivial algorithms introduce artifacts. The trick is to choose an algorithm such that the artifacts which are introduced do not interfere with the information you wish to enhance.

This is most often used for artistic purposes and not information.

> Authors in image processing jourmnals often select test images to make their algorithms look better than they are. Here is a test image that I selected to bring out artifacts, other than noise, in algorithms for increasing contrast, > > http://www.general-cathexis.com/Manual/CarKabob.png > > Here is the result from applying the CLAHE JImage plugin with default settings: > > http://www.general-cathexis.com/images/CarKabobCLAHE.png Wonderful artifacts; very clear. > Here is a result of my proprietary "hysteresis" method: > > http://www.general-cathexis.com/Manual/HysteresisContrastEnhanced6.jpg More subtle artifacts, particularly the difference in the sky either side of the left-hand (cellphone) pole.

The problem is with enclosed and partially enclosed spaces. The purpose of the example is to educate users to predict which images have problems.

> Also, the saturation in the image appears to increase;

Actually, that is an intentional increase in color contrast which has some unusual artistic benefits. Without adjacent areas of contrasting color, the color saturations wouldn't change. I can easily avoid increasing such color contrast by only processing the Y channel of YCbCr, V channel of HSV, I channel of HSI, etc. color space.