Undoing halos and 'chroma fogging?'

[tomczak]

Here is a crop of an image showing, for the lack of better name, 'chroma halos/fogging' (e.g. edge of orange lifesaver). I'm not sure, but to my eye it seems to be a combination of low-pass sensor filter, chroma noise reduction and USM-like sharpening done by the camera. The question is: if that's the only version of the image that one has, how to put it back in the bottle, preferably automatically?

I was thinking that, for example, since we know how USM produces halos (i.e. making a blurred copy of an input image and than darkening pixels darker than their blurred version and lightening lighter ones), can the reverse be done (e.g. lightening pixels darker and darkening the lighter than their blurred version), and thus hopefully reducing the halos. I can't figure out a procedure that could work.

This may seem an esoteric problem, but it happened to me several times where I had to rescue images that either I, someone else or the camera oversharpened before, and I still don't know how to do it gracefully.

[tomczak]

I've tried several ideas and one method that seems promising at somewhat reducing already existing USM over and under shots at edges is to Composite in Soft Light mode the input image with a Negative of High Pass filtered version of it.

This is less effective on 'saturated colour leaking', such as the lifesaver edge - I imagine because of the chroma noise reduction smearing that may have contributed to creating it in the first place.

[den]

The image sampling is really small to know if this will be of help or not....

(1) Apply a Mask Tool-Brightness Curve 'BrokenLine' = [0,0], [5,0], [20,40], [50,100], [80,40], [95,0], [100,0] and then apply a small blur [Blur = 1-->3 depending upon image pixel dimensions, usually 2 for a 3456x2304 image] to the image...
(2) Use your favored 'local contrast enhancement' technique, for example for the posted image sample, USM settings: Amount=50, Radius=40, Threshold=0...

This will not address the chroma spill overs but the mask will limit 'local contrast enhancement': (1) to mid-tones without additional deep shadow or high highlight pixels to decrease detail and (2) the nearly 45 degree slopes to/from the [50,100] control point tend to favor the darker half of the contrast enhancement rather than the lighter half.

[tomczak]

It seems that the 'colour halos' can be isolated to HSV-S channel. How to reduce them? This is fairly common, from what I've seen.

[den]

Maciej...

I have been trying off/on these last few days to find a non-manual technique(s) that would be generic for most images but cannot find the right workflow...

(1) the Composite-SoftLight blending of a 'negative' HiPass image does lower the contrast at edges but it is difficult to only limit its effect to the pixel width of the 'halo-ed' edges without lowering the contrast in other image areas... even with multiple width edge masks...

(2) the chroma halo about the 'life saver' certainly can be masked but the replacement for this image area needs both color/tone/texture... to be realistic

By the time one tries to isolate image areas, apply masks, and apply a variety of transforms, I seem to do manually in under five minutes the following...
(1) create a 'color negative' of the image to better show color halo... notice that there is also color spills in the lower middle of the image, the lifeline chrome bracket, the navagation light fixture, and along the right side cabin side/top and the foilage...
(2) use the Paint Tool - HSL Hue & Saturation Only to correct the identified image areas hue and saturation, sampling preference colors from nearby image areas...
(3) transform the resulting(2) 'color negative' image back to a 'color positive'
(4) adjust midtones for contrast...

IMG_4024_crop-1b.jpg (25.78 KiB) Viewed 11356 times

[tomczak]

Here is another example of fogging. Since the 'fogging' was mostly limited to the Red channel, I isolated RGB-R channel, USM-sharpened it with the radius similar to the extent of the 'glow', then Composited the original and the sharpned RGB-R in Darken only mode (I only wanted to lessen the glow, not sharpen the edges). Finally I recombined all 3 channels, replacing Red with it's processed version. The glow is somewhat less offensive to my eye, but still there.

I'm trying to draw some general conclusions from it. What causes glowing? Antialiasing filter? Poor lens? Some hidden chroma noise reduction algorithm?

Then is it always going to affect the Red channel more then others? Why? Is it specific to the camera or caused by some fundamental phenomenon?

p.s. would it be possible to label the channel windows so that one knows which channel is which (e.g. RGB-R, or HSV-S etc.) - it's really easy to get lost if there is more then a few channels and the operation is somewhat complex.

[tomczak]

I've seen similar 'blooming' in other pictures. The 'fog' is almost always reddish. I wasn't sure whether it's some sensor-specific blooming or RAW processing artifacts, but here is an idea that I begin to lean towards - its the diffraction affecting the longer waves more than the shorter ones. Could others comment on that or their experience?

http://www.cambridgeincolour.com/tutori ... aphy-2.htm

I was also thinking of a method that could reduce the effect. The best idea I have is to sharpen R channel more than the other two (perhaps using the Darken trick first). Is there a better way (other than by hand), especially if it could be done in a workflow?

[tomczak]

Could someone offer a physical explanation while such edge 'fogging' occur? This time is blue on green... Cheers!

[HanSch]

I have no physical explanation, but it looks similar to effects that I observe in images of a Lumix camera, when the optical I.S. has to work hard. Mainly long focus distance and pretty long exposure time.

[den]

Erratic IS would not have occured to me... ...excellant suggestion for investigation.

Regarding those using Canon IS lenses on tripods, this may be of interest: http://www.dlcphotography.net/TripodAndIS.htm