Perception Based Contrast Enhancement by Majumder


#1

Hi,

I recently came across this paper by Aditi Majumder:

It’s an interesting read and the idea seems simple and logical (I’m talking about understanding here, not coding… ). She also seems to share the source code here:
http://www.ics.uci.edu/~majumder/contrastcode/codecontrast.html

Does Anyone know if this technique is implemented in any FOSS?
Cheers


PhotoFlow News and Updates
(PkmX) #2

Interesting paper. The idea is pretty simple and the algorithm does not seem to be difficult to implement.

I see one artifact with this approach however, given “Lenna” in greyscale:

(left: original, middle: this paper with τ = 0.5, right: imagemagick contrast +15)

Notice that the bow-shaped part enclosed by her hair at the bottom of the photo becomes much brighter than the body after contrast enhancement. This closed area becomes a separate hillock during sweeping and is stretched more than other parts due to a looser constraint (6). This becomes pretty noticeable with increasing τ. With that said this paper does a pretty good job at enhancing the texture of the hair.

I suppose this approach should work well on images with high contrast as it avoids over-saturating pixels, but then I have no idea how well it compares to other tools available in darktable/RawTherapee.


#3

Yeah also the nose is enhanced quite strongly. But let’s put it this way, I could live with some larger, easy to separate areas where some “artifacts” are created if I get some nice local contrast in the complicated parts like the hair, the feather and the hat. Repairing these “artifacts” is not that hard.


(Pat David) #4

I wonder if @David_Tschumperle has seen this also?


(Kevin Payne) #5

Isn’t C2G in GIMP perception based contrast enhancement?


#6

Oh interesting I hadn’t noticed that…though the algorithm seems to be different…