Cornish Pic

Matt,

I like that as an image, but not as a photograph. It is better than the examples on the HDR site that Lilly linked to, but since I like photos to look like photos, and not like the product of a paintbrush or airbrush, I can't ever see myself using this technique.

G

George,
Fair comment I was in a "graphic" mood at the time

Of course, you can apply an easier technique by taking one or two shots, 2EV apart and manually blending them in Photoshop with layers. Saves lugging around filters (which are expensive with the 12-24mm).

In fact, most of the tone mapping can be done manually in Photoshop - and I have a shot that just won't tone map properly. May have to resort to manual blending anyway

Now on a similar topic... I really, really want Pentax to add full 22-bit RAW to the K10D and Silkypix. That'll add some serious dynamic range and save most of the messing around with HDR so it can still be a "photo"

Matt

Hi all

Excuse me for being thick but what is HDR and what are you guys doing to create it what ever it is does it envolve software


Pete

HDR = High Dynamic Range
HDRI = High Dynamic Range Imaging.

It involves a few things:
1. Taking a series of bracketed exposures (on a tripod). Say, -4EV, -2EV, 0EV, +2EV, +4EV (that's four stops underexposed, two stops underexposed, "correct", two stops over, four stops overexposed)

2. Combining the set of images in software into an HDR image file. This is 32-bits per pixel (ie 16million times greater range than a JPG). The HDR image contains a wider range of dark to light than can be represented on any monitor (e.g. the shadows to the brightness of the sun)

3. Doing any adjustments on the HDR file

4. Then mapping the information contained in the HDR file down to a "normal" image. This is called "tone mapping" and is where the "unrealistic" component creeps in.

Here are two "pro" examples:
http://www.ephotozine.com/gallery/showlargepic.cfm?photoid=804864
http://www.ephotozine.com/gallery/showlargepic.cfm?photoid=814571

Note that you could never normally obtain both the streetlamp and the shadows... not even with a filter

It's complex, tricky, and at the moment a complete pain in the neck!! But I find it's an interesting challenge...

Hope that helps!
Matt

Thanks Matt,

I think i have read in one of my magazines about this it can make for some excelent results.

Is there any software for doing this you can recommend


PS can you direct me to any before and after pics to see a start to finish example ?

Thanks

Pete

Pete,
Well, not exactly recommend. Personally I don't think any of the software is 100% right now. But to get you started here are a few links:

Artizen: http://www.supportingcomputers.net/ - good, but excruciatingly slow, even with a seriously fast machine.

Photoshop CS2 - has HDR support. Fairly poor at combining the images unless they are 100% the same (just different exposures). Artizen is much better at assimilating the images. Tone mapping is pretty good, but takes a lot of practice.

Photomatix: http://www.hdrsoft.com/ - Their tone mapping isn't as versatile as Artizen. It's a touch faster, and combines images well (even with people moving etc). Didn't like my K10D DNG files though - produced some very strange artifacts! Their website has some examples (which I don't like personally!)



Just bear in mind that most of it is very experimental. There are many, many tone mapping algorithms available - all work with different image types. The other thing is that we perceive prints differently compared to web images - so an HDR print may look really excellent, but a bit "neurgh" on the web

Matt

IMHO 22-bit is a kinda marketing trick. May be it has 22-bit internal image processing but at the end you will see the same DR like k100 has. If you want an evidence of that just look at my previous pictures. They are not just JPEG from camera but RAW mapping to JPEG. IMHO you can take from RAW about one and a half of stops from dark area and a half of stop from bright area. If you can do more than that let me know.

Not exactly
Pentax cut down on the number of stages from the CCD to the ADC. Normally you'll have an amplifier to set the ISO rating between the CCD and ADC. From what I gather, Pentax have no intermediate amp and so the ADC sees the full dynamic range all the time (ie all the ISO ratings).
When you set the ISO rating, you specify which range of values should be taken from the ADC and saved in the 12-bit RAW file.

So, if you had a "full" RAW, then you'd cover a much larger dynamic range You would expose for the highlights (ETTR), but the 22-bits should give a much deeper shadow range. (But you'd have problems with processing it, and you'd still need to tone map it )

Your estimation is probably around about right with the current RAW files. I reckon I get a bit more out underexposed... but maybe only another half stop. There's no excuse for at least trying to get the right exposure
Matt

I don't know how good you are in electronics but it sounds not good for me. I've never use CCD element but I know about ADCs (analog-to-digital converter). There are two reasons I have doubts about:
1. I have no idea how it's possible to convert an analog signal to digital value with that precision. There are problems with noise at reference voltage, ground level and etc. at least camera is a portable thing and you can bring one wherever you want where temperature is 0..40 (unofficially -20*C is not a limit).
2. The most efficient way to estimate a value is to fit the signal into input ADC range as close as possible. In this case the estimation would be most accurate. Amplifier is not a problem. I guess the sensor could be programmed for ISO and most of the work is already done there but I have no information about type of the sensor.

I have a theory it could be. The arranged signal goes from sensor to 22-bit ADC input but the accuracy is not 22-bit good then pentax make nonlinear conversion and fits the values into 12-bit output file having acceptable noise level.

True, you don't. I know my electronics I shalln't bore you with the projects I've worked on, but believe me...
e.g. http://focus.ti.com/docs/pr/pressrelease.jhtml?prelId=sc05240

I based my assumptions on:
1. The fact that 22 bit ADC are not a problem at all in modern electronics
2. There was a video interview/transcript with one of the designers/managers of Pentax Japan who said they were aiming at reducing the number of electronic stages between CCD (which is analogue) - hence my comments about removing the amp (at least simplifying it).

At the end of the day it doesn't matter at the moment... the K10D produces a wonderfully clean image. But I reckon that a larger RAW file will happen sometime (even if it's "only" 16-bit)

Matt

Finally, it's good to see a person who understands you.
I guess it's not the chip pentax could use but at least I can explain my doubts using numbers.
Now about your link. We have digital camera

dimension of frame =3872x2592
3 base colours
about 3 frame per second

That means we have to be able to do 90`326`016 measures per second. Using this chip we have two options:
- 90326/125 = 722 chips in fixed channel mode
- 90326/23.7/16 = 238 chips in automatic mode and 20-bit resolution

I think both variants are not acceptable for portable device.

Now about precision. For example we can use 5V input range and 24-bit resolutions means we have 0.3uV as a minimal value (l will call the value as step). This converter has AD conversion noise = 12uV (40 steps - 5-bit of resolution) and
temperature drift = 0.5uV/C (40*0.5 = 20uV or 66 steps - 6-bit of resolution for overall temperature drift)

That's why I am so skeptic.

IMHO that could be a good idea but I don't think we can expect true 22-bit precision at any processing level but may be this approach makes the noise more natural and more appropriate for human eyes. But for marketing it's the really big break.

k10 is great (that's why I bought it) but it's not 22-bit great and I knew it.

(I plucked the 24-bit ADC out of Google just to show that 22-bits isn't really a big deal, even at data rates. BTW, I seem to remember that the larger CCDs have multiple modes where they can bucket-chain out the values in 2 or 4 channels to increase data rates... but that may not be on the K10D)

Beg to differ on this point - the CCD is actually 3872x2592 light values - not RGB pixels. The pixels are arranged in a four square bayer matrix (R/G/B/G). So, the figures are a factor of 3 too large The RAW conversion (either in camera or on PC) has to demosaic the image to get a full 3 primary colours per pixel.

I certainly agree that the lower bits of the ADC are almost certainly just noise (and it won't be 5v, most certainly 3.3v).

However, think about it this way - assuming that there is no variable gain pre-amp between the CCD and ADC - there only needs to be a 5 bit range over and above the image range. So, assuming the usable image range is 12-bits, then 17-bits will cover 12-bit image over a 5-stop ISO range. (Since each ISO step is a doubling in intensity, and therefore one bit in binary.)

My point was, not that I'd need all 22-bits, but they might as well create a RAW format that dumps the lot anyway so the software can make decisions about the noise floor and we'd get a near-HDR single image that covers at least 5-stops more than the current single RAW frame

(Like other manufacturers you can always "gang" the pixels together for the very low light and reduce the effective pixel count.)


So, we'll just have to wait and see
Matt

It seems like we are talking the same things by different words.

IMHO true 22-bit ADC is not a problem only in laboratory (i mean result, not just a fact it is). My calculations of the chip was just an estimation of approach more so we have no information about the sensor.