Examination of the black dot issue...example

Started Dec 11, 2008 | Discussions thread
Flat view
RicksAstro Veteran Member • Posts: 3,670
Examination of the black dot issue...example

I'll start by saying, if you care about just making beautiful pictures, ignore this thread...it'll just frustrate you. This camera creates some incredible images!!!

But if you're a gearhead geek like me, or are interested in astrophotography, this may be of interest...

I wanted to see for myself what's going on with the completely raw values, so I took one of my xmas tree images from a few days ago and looked at it using my astro SW (MaximDL) which allows a good analysis of images.

I took DCRAW's undemosaiced 16 bit TIF and ran it through to see where the black dots pixel levels were at. They seemed to be between 290 and 996...when I set the threshold to 996, ALL pixels below 996 were immediately to the right of very light areas. I saved this threshold image (black dots were completely black, all else completely white). Remember, this was from the undemosaiced RAW file.

I then demosaiced this image in MaximDL using it's high quality algorithm.

I took both these images into Photoshop to examine the correlation of essentially dead black pixels and the demosaiced image. In some cases (where there was only a single dot), the demosaicing algorithm blended it all in. For the larger blocks of pixels, it showed up more in the real image. I made a blinking animation showing the black dots as white over the color image. This is a 100% crop of the original:

My observations:

  • No color channel is immune...there were larger than 4x4 blocks of raw pixels.

  • It's not just one or 2 pixels that are dedl....in this example, there are many hundreds of pixels which were below the black level of the image!!!

  • The effect on the final outcome is relatively minor in comparison to the number of pixels involved

  • Lighter backgrounds don't show it as much as darker backgrounds.

One way I could see Canon fixing this is to detect these "below zero" pixels and to take the median value...not the best solution since you'd be losing some data, but for all intents and purposes would satisfy even the biggest pixel peeper.

My intent is not to make this into a big problem, since I don't think it is. It's just my engineering curiosity getting the better of me.

If you're into taking actual photos rather than actually looking at pixels, I TOLD you not to read this...


-- hide signature --
Flat view
Post (hide subjects) Posted by
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow