How to fix purple noise caused by a too-bright sky

[redblue]

Hi,

a question for the filter experts:

I'm looking to fix videos filmed with a Pentax W90 digital camera.
This is a rugged waterproof camera great for filming sports
activities where normal cameras would not survive. But the
videos have serious brightness problems:

Whenever there's too much bright sky or sun in the picture,
I get vertical areas filled with purple noise.

Is there a filter that can find the areas and darken the purple
noise?

Below are some sample frames with typical problems.

http://img833.imageshack.us/img833/4346/pentaxw901.jpg (http://img833.imageshack.us/i/pentaxw901.jpg/)
http://img231.imageshack.us/img231/1792/pentaxw902.jpg (http://img231.imageshack.us/i/pentaxw902.jpg/)
http://img832.imageshack.us/img832/6175/pentaxw903.jpg (http://img832.imageshack.us/i/pentaxw903.jpg/)
http://img21.imageshack.us/img21/9201/pentaxw907.jpg (http://img21.imageshack.us/i/pentaxw907.jpg/)

Any suggestions are welcome.

[scharfis_brain]

This is CCD vertical smear combined with a bad IR-filter built-in in the camera.

the purple fog is caused by infrared light being passed to the sensor.

You cannot do anything against it in postprocessing AFAIK.

[2Bdecided]

Tell everyone that it's a special effect you worked very hard to achieve, and it's supposed to look like that!

If scharfis_brain is right, try adding an IR blocking filter in future. Be careful when purchasing: "IR filter" can mean something that blocks visible light, but lets IR through - you need the opposite ;) ND filter might help too.

Cheers,
David.

[redblue]

Thanks for the fast replies.

I might actually try taping a UV filter from a bigger camera to the outside of the camera for a test. It's not totally impossible.

This is the camera, by the way:
http://www.amazon.com/Pentax-W90-Waterproof-2-7-Inch-Black/dp/B0039237GU

This is CCD vertical smear combined with a bad IR-filter built-in in the camera.

the purple fog is caused by infrared light being passed to the sensor.

You cannot do anything against it in postprocessing AFAIK.

I'm OK with the 100% white CCD smear - nothing to do about that. But the purple itself?
No way of turning the purple parts of the frame into a mask and use that part to change the
contrast of the frame to darken the purple bits?

This is how far I got:
I can see the purple areas that need fixing in this script.
I haven't managed to replace them yet.
For the example video1 is the clip that needs fixing (I use the lower resolution as a test)
video2 is a grey clip I copy the color info into.
This has only the color info. Here I look for purple.
video3 is 100% green. As a test I want to replace the purple bits with the video3 areas - later this will be a darker black and white copy of video 1.


video1 = DirectShowSource("f:\dcim\104_0408\IMGP2114.AVI"). KillAudio.trim(5196,7596).ReduceBy2
video1 = video1.ConvertToYUY2.crop(0,0,180,240)

# Grey
video2 = BlankClip(video1,color=$7f7f7f)
# Green
video3 = BlankClip(video1,color=$00ff00).ConvertToRGB32


video2 = MergeChroma(video2 , video1 ,1).ConvertToRGB32

video2 = ResetMask(video2)
video2 = ColorKeyMask(video2, $B26A91 ,43)

video2 = Layer(video3,video2,"add",255)

return StackHorizontal(video2,video1.ConvertToRGB32)



This works at identifying the most of the purple areas, but
has some false positives.
The approach ignores also the information, that the purple
areas are vertical stripes.

[scharfis_brain]

UV light isn't the issue causing the purple effect.
It is IR light that creates the purple appearance.

[PhrostByte]

You can see the same purple effect with most cheap cameras -- take a picture of an IR LED with a phone and you'll see it.

I don't think you'll be able to filter that in software without losing some real shades of purple, but that seems to be a rare enough color that it might not matter.