Talk:Demosaicing

From Wikipedia, the free encyclopedia
Jump to: navigation, search
WikiProject Color (Rated Start-class, Low-importance)
WikiProject icon This article is supported by WikiProject Color, a project that provides a central approach to color-related subjects on Wikipedia. Help us improve articles to good and 1.0 standards; visit the wikiproject page for more details.
Start-Class article Start  This article has been rated as Start-Class on the project's quality scale.
 Low  This article has been rated as Low-importance on the project's importance scale.
 

Color filter array[edit]

In the first paragraph, I replaced color-filtered by color-filtered because the Bayer isn't the only color filter. I hope somebody will start Color filter Array if ther is enough demand.--Marc Lacoste 11:35, 13 November 2005 (UTC)

proposed merged[edit]

I think Bayer filter#An illustrated explanation of color image reconstruction is more interesting here, Bayer Filters are less generic. --Marc Lacoste 23:41, 19 March 2006 (UTC)

Support

  1. Marc Lacoste 23:41, 19 March 2006 (UTC)

Against

Done --Marc Lacoste 10:22, 15 April 2006 (UTC)

Now at Demosaicing#Illustration.

Scaled Images render unpredictably due to browser antialiasing - I've set all to 300px
It would be clearer if 'reconstructed' were not antialiased - maybe another image 'demosaiced' and 'antialiased'?
For control the Original+Reconstructed images should be scaled without antialiasing - they are 100px now.
---19S.137.93.171 (talk) 05:06, 8 May 2012 (UTC)

Actually, I suspect shenanigans. I totally distrust that illustration. It expands 1 input pixel into 9 photosensors. The Bayer mosaic has 4 photosensors per pixel. Is it subtle Foveonista propaganda ? Can someone re-do it realistically ?

For more realism we should implement an Anti-aliasing filter that blends each pixel into 4 pixels first !

---19S.137.93.171 (talk) 05:22, 8 May 2012 (UTC)

Explains why digital cameras are less good than they ought to be[edit]

I'm adding the following here, rather than the article, because I can't write it authoritatively. Hopefully someone else can re-word. The basic problem with digital cameras is that they are actually pretty hopeless! Consider viewing the output of an 8 Mpixel camera on an LCD monitor, enlarged to "1 pixel per pixel", i.e. 1 px of the image maps to 1 px on the display. The result ought to be perfect at this scale, but it is invariable poor, with either severe blurring, or very obvious colour fringing. Essentially, one ought to take the quoted Megapixel resolution of the camera and divide by 9 to obtain a resonable "truthful" resolution. (Actually, 16 would really be necessary, but image-processing can help a bit. Nevertheless, image processing cannot recover the really hight frequency spatial data which has already been lost.)

The authoritative fact is that you have been misled if you believe that a digital camera picture should appear "perfect" when displayed "1 pixel per pixel". The article does explain why, I think, but your factors of 9 and 16 are just made up exaggerations of the problem. Just stick to the reality instead of the exaggeration. Dicklyon 07:11, 13 December 2006 (UTC)

The thing is that none of the discussions of digital cameras contains any of the theoretical constructs of Digital Signal Processing. Considering the Nyquist limit, it is necessary for a digital camera to have a "low pass filter" to slightly blur the image before it reaches the sensor. Such filters would be ideally defined in lines -- a line being two pixels. If we consider an ideal filter with line size of twice the pixel size -- such a filter would blur such a line pattern to a uniform color that was the average of the bright and dark lines -- this filter would eliminate all aliasing for a monochrome sensor.

However since the color resolution is based on 2x2 blocks of pixels, this camera will still record color aliasing. It needs to be understood that once this aliasing is recorded, there is no mathematically correct method to eliminate it. So, to eliminate aliasing, it would be necessary to use an anti-aliasing (low pass) filter that removed lines less than 4 pixels.

So, I have to say that your division figure is based on an incorrect assumption. The fact is that to determine the line resolution you would have to divide the pixel resolution by 4. So, a proper anti-aliasing filter (theoretical filter) would only divide this by 4 again. So, your figure of 4 x 4 = 16 is correct for line resolution from pixel resolution but the CFA only reduces resolution by a maximum of 4.

This reduction in resolution is somewhat confused by the fact that the Green resolution is greater than the Red or Blue resolution. However G resolution it is not twice the R or B resolution but rather it is √2 times the R or B resolution. So, the question is whether the anti-aliasing filter should be based on the total color resolution or on the G resolution.

IAC, the resolution of the camera is going to be limited by the anti-aliasing filter and a filter that allows smaller details than the nyquist limit to be recorded is going to record artifacts that cannot be removed in a mathematically rigorous manner.

I also find it odd that DSP methods (FFT and/or digital filtering) are not used or discussed as methods for demosaicing. I should be clear on this that the math of DSP is clear that interpolation DOES NOT WORK. Tyrerj (talk) 17:53, 1 June 2012 (UTC)

Simple explanation of colour fringing[edit]

Here's a way to explain colour fringing - it really needs a diagram. Imagine photographing a test object where the top-half is white, and the bottom-half is black. Then, the horizontal dividing line will lie over the mosaic in some way, dividing the sub-pixels. In the worst case, you will get an image with the top-half white, then a row of blue+green = turquoise, then a row of red + green = yellow, then the bottom half black. This is most unpleasant! It's even worse with purple fringes.

It's a good thing that people who develop demosaicing algorithms don't just go with the first simple thing that occurs to them, huh? Still, some problems are inevitable, since information is missing and needs to be "made up". Dicklyon 07:13, 13 December 2006 (UTC)

Your explanation would be correct except that a camera has an anti-aliasing filter which blurs the image slightly. In this case it would blur the edge between the Black and White areas. Tyrerj (talk) 18:06, 1 June 2012 (UTC)

Figures missing[edit]

Some of the figures that were used in the article were deleted from commons, apparently for lacking a category. Then the refs to them were removed from the article, leaving extra blank lines. We can probably do fine with the images we have, with editing to patch it up. Dicklyon 21:39, 30 November 2007 (UTC)

Spelling[edit]

Demosaicking is acceptable, but demosaicing is about twice as common in books, according to a quick search. So I reverted the move and respelling based on the unsupported assertion that the k is required. Dicklyon (talk) 15:51, 13 November 2008 (UTC)

The problem is that these books are written by people familiar with the process of demosaicking but not necessarily at all familiar with spelling in English. Words like picnicking, politicking, and bivouacking -- in fact any -ing form of any verb ending with -ic -- is spelled with a k between the ic and the ing. (The only exception I've ever heard of is sicking, which has the equally valid variant siccing.)
In fact I just searched in what is by far the most extensive word-search utility available on the Web -- at the National Puzzlers' League website -- where the list of searchable words is consolidated from a number of large lists (including the Merriam-Webster Second International unabridged dictionary, the Merriam-Webster Third International unabridged dictionary, and the Official Scrabble Players Dictionary). The total number of entries is over 500,000. Among *all* words ending with -icing, *none* of these come from verbs ending with ic.
Not one.
In particular, the non-technical word "mosaicking" occurs in these lists (as it does in many respected dictionaries). By contrast, *not one* of these "unabridged" lists (or respected dictionaries) has "mosaicing" with the -icing ending.
And surely it makes sense that "demosaicking" should follow the same spelling pattern as "mosaicking".
My point is that Wikipedia should not just parrot someone else's mistake. Instead, if there is a solid case for holding the line against a mistake -- and I believe I have described a solid case for why "demosaicing" is a mistake -- Wikipedia should take the high road. It should not help propagate mistakes.
(For what it's worth, I am a language professional: for twelve years I have edited the crossword puzzle in a leading U.S. publication.)Daqu (talk) 03:09, 18 November 2008 (UTC)
That all makes sense, sort of. If you can get a consensus here that the common spelling is simply a mistake of illiterate techies, then we can move the article and change the spelling. But you really to need to get a consensus, not just do it. I admit I took a similar position to yours on Fitts's law, but it was also backed up by historical precedent (that is, the correct singular possive Fitts's was once more common than the now common Fitts'). Dicklyon (talk) 05:40, 18 November 2008 (UTC)
I agree with Daqu, the title of the article should be "Demosaicking". I think the correct spelling is "demosaicking" for all the reasons Daqu states (and especially because of the argument that it should follow the spelling of "mosaicking"). That leaves the argument about whether we should use the correct spelling or the most common one. For "demosaicking", which is a jargon word, I think we should use the correct spelling. (For words in more widespread usage, especially words about popular culture, I might be persuaded to use the more common spelling). There is also the question about which spelling is indeed more common - I don't think we can say for sure which is indeed more common, anyway.
Finally, as an aside, I've read quite a few articles on demosaicking recently and there is fairly widespread use of both spellings. I've even seen different papers by the same author using different spellings, and papers where the spelling of demosaicking in the url differs from the spelling in the title of the paper.
Martin.Budden (talk) 20:51, 6 January 2009 (UTC)
It should be ‘de-mosaicking’. — Chameleon 16:17, 13 September 2009 (UTC)
Are you saying you don't believe my estimate that "demosaicing is about twice as common in books"? Or that it should have the k for some other reason? Dicklyon (talk) 00:15, 14 September 2009 (UTC)
I don’t care how books, written by people who are technically strong and linguistically weak, commonly spell the gerund of this new verb. Similarly, tech nerds think that the agent noun from ‘refer’ is ‘referer’. They also omit the space in the verb ‘log in’, even though this would imply ‘loginning’ as the gerund/participle instead of ‘logging in’. The Commodore 64 kernel was internally misspelt ‘kernal’. English has well-established rules of spelling and grammar. Tech geeks can be ignorant of them if they like, but not impose them on the rest of us. — Chameleon 01:21, 26 September 2009 (UTC)
So you're saying that if we report the most commonly used spelling, we, or tech geeks, are imposing something (unclear what "them" you refer to) on "the rest of us"? Interesting. Dicklyon (talk) 04:30, 26 September 2009 (UTC)

Define your audience/Define your terms[edit]

This article does not define terms such as "incomplete color samples" and "spatially undersampled color channels" . Thus the article will is only be readable by people already highly knowledgeable on the subject. ie those people who do not need to read it! I had to look elsewhere to understand what Demosaicing is and I would suggest explaining it in simpler terms such as: "Each pixel can only output a value for one of the primary colours (Red or Green or Blue). So the actual colour must be calculated using additional information, such as the colours of the surrounding pixels. This process is Demosaicing" See? Simple!