shot from earlier today
| Photographer |
spent the afternoon shooting a Pug
| Photographer |
Oh that is just adorable Bob. Pugs are so much fun, Mastiffs at heart but still scared of their own farts.
My only criticism would be the collar sticking out of his left ear, it took me a minute to work out what on earth it was. I'd clone out the distraction if it were me.
| Photographer |
he was a pain in the butt to photograph under studio conditions, but he is a handsome chap
| Photographer |
kalniel (29-05-2013)
Nice Robert. Do people pay to have pictures taken of their pets?
To err is human. To really foul things up ... you need a computer.
Hehe, spoken like a true non-dog person.
Yes.
Yes, they do. And often quite a lot. I remember many many years ago, at some point when I might have even still had some hair, my wife had some pro studio shots of our cavalier done. This was before I was 'into' photography and let's just say the cost of the framed photo was enough to spark a landmark argument.
That photo is still on the wall and always will be, even though I could and have taken better shots (if I do say so) we still love it.
| Photographer |
I only ever met a pug once who belonged to some kids I was shooting their grand parents' diamond wedding, but hugely entertaining, providing you don't mind the litres of saliva dribbles all day long!
In theory, yes - if you've captured information +- 2 stops then you've got an additional 4 stops of information in there. The question is how much extra information you have in the RAW vs the jpeg - which depends on the camera - it's *possible* that you have the same, or more, or less. Can't tell you how that particular software would compare though - try it and see I guess.
Interesting point. I reckon kalniel's right - it will depend on the camera.
JPEG is 8 bits per pixel, i.e. 256 different levels of brightness for each of the primaries.
RAW is up to 14 bits per pixel, i.e. 16,384 different levels of brightness for each of the primaries.
I doubt peak white on JPEG = peak white on RAW as there's just more headroom in the latter. But where exactly - in other words, how many of the bits are over peak JPEG levels, I don't know.
It's really not like that.
So RAW is as the name suggests, the RAW data, its compressed in a lossless manner. For most of us, this means 14bits per channel. The channels are in groups of 2x2, with 2 green, 1 red, 1 blue. This is due to the human eyes love of green.
A JPEG on the other hand is a bit different. It exists in the fourier domain. This is an inherently lossful process, it is based around the idea of repetition of signal frequencies, to this end JPEGs are divided up into squares, which each describe their own type of data, then subdivided further and so forth. You can really see this effect if you massively compress a JPEG. Try it, pick a photo you like and turn the quality down in something like Paint.Net which gives you a live preview. Notice how details stay preserved where there are lots of them in that little square? Yet if there is only 1 detail (say a single star in a black sky) it's preservation is different to if there are many (ie, many stars in a black sky).
This is why I had to jump on this. A jpeg IS NOT 8 bits of information per pixel. Not at all. Lossless JPEGs maybe, but they aren't true JPEGs. It is 'Gamma Compressed'.
When you are fitting curves to discrete values. It is the detail that is compressed, not the range. It is also the local detail. Comparing 14bits to 8bits is really incorrect. It depends on the distribution, and how that distribution of levels ties up with the sub-squares jpeg uses. Sure it ends up as 8bit RGB, but so does the RAW you look at.
So keep in mind, that the JPEG doesn't lose range, or even fine detail, it loses overall detail, based on rate of change.
But lets go back to the original description of a bayer filtered RAW file. It has more green. That means that the green is normally used for the luminance (you can think of it as brightness) and the Red and Blue are used to determine which ends of that spectrum the luminance should live in. Obviously, this is flawed. If you are taking a picture in an environment with lots of Blue light, but no red or green (cold pure blue if you will) how will it work?
Well a quick shifty at this graph:
http://t3.gstatic.com/images?q=tbn:A...enVH51TBul4SWg
Which shows us the frequencies the human eye sees, shows the level of the problem (ignoring rods and cones at this point).
Look how far away that blue is! If we rely on the Green component for the luminance, we aren't going to get a proper response. We will end up with a very over-compensated amount of Blue light.
The reason I mention this, is simple, RAW processing is bloody complicated. Full of many rules, tweaks, and fudges to get the very best of the data available. If we decide to transform our image to a lossless bitmap we irreversibly cut out that data. It is this action of developing the RAW image where the loss occurs. It is when the assumptions about what the RAW data should be and what the noise is.
A lot of information is lost, not in the change of 14bits to 8bits (as often +/-2bits of the data are useless anyway) but in the conversion to colourspace as we know it. This is another really important thing to think about. What is colour? CIE 1932? We have to map these filtered channels into a colour space that makes sense, this is before we even get to think about white balancing. When we map these sensor readings to the colour matrices (such as sRGB or aRGB) we are, once again, going to be losing precision. If you've got something thats full of green, you will get different local sensitivities between aRGB and sRGB. Again, this is before a reduction in 'bittage' has happened.
So very rough re-cap, of what you 'loose' by having JPEG not RAW.
- Luminance and Croma losses are made by assumption of Bayer to Pixels. (this is due to the 'smoothing', to make each bayer cluster, behave like a pixel)
- Gamma Space is compressed, irreversibly
- Images are often 'sharpened' to compensate for step 1, the logic used might make your image noisey.
- Images are often de-noised to compensate for above step and general sensor behaviour. Some is done in step1 too.
- Image is clipped to a Colour Space. Data is lost in the conversion here.
- Image is downsampled to fit in JPEG, this is the 8bit conversion, but given all the previous steps, the loss is not really 14bit to 8bit. Mostly due to colour space conversion.
This colour space conversion is so lossy, because effectively we are dot by
0.6097559 0.2052401 0.1492240
0.3111242 0.6256560 0.0632197
0.0194811 0.0608902 0.7448387
If that is with 14bits of precision, we've just lost a hell of a lot of data haven't we. Far more than dropping to 8bit later will do.
TL;DR Use RAW. Colour and our eyes are more amazing than we take the time to think
throw new ArgumentException (String, String, Exception)
But as you point out - information isn't the same as range. You lose a lot of information in jpeg, no-one is questioning that, but do you lose more range via sensor clipping in one exposure?
If there is information beyond the range of the sensor to record at a metered 'correct' exposure then it's lost, regardless of using RAW. But if you bracket exposure you can capture it.
Well, it really depends how your bracketing! If you are using sensitivity bracketing, then you will find +/- 2fstops on most sensors gives you more information. Very, very, very, roughly, most sensors have a range that is around -/+1.5 fstops, however this depends really on the levels of light, and some sampling theorems, it isn't just a simple nyquist rule. However, as I mentioned when it comes to deciding the luminosity, which is normally what we are bracketing to get (as chroma should be constant) it really does depend a bit. The local swing for each cluster will depend on the image processing algorithm. This is important to consider because some set of pixels will have more 'sensitivity' than others, this is because each bit of the sensor will behave very slightly differently.
From your question I guess we are talking about bracketing to capture more information. This is because some parts are burnt out, others under-exposed. So basically the dynamic range can be captured in JPEG just as well as it could be in RAW, if the image processor gets it right. Some DSLRs have a dynamic range mode, to tune this functionality, it will give better shadows most of the time.
Then we come to the conclusion I really don't want to say, if you've not got it from your JPEG, you should bracket, but you shouldn't really be shooting JPEG anyway, as you have no option to tweak the logic used to translate the colours from the raw data. You've lost a lot as is.
It's not a helpful conclusion I grant, but I think its better people really do think about how the cameras work.
throw new ArgumentException (String, String, Exception)
There are currently 2 users browsing this thread. (0 members and 2 guests)