Anyone feel that there is no difference between the two images one with photomatix and one without?
I'm not trying to bite anyone's head off. I'm just looking for opinions and facts. What actually happens to each individual pixel when you do use photomatix or other blending techniques? Are pixels damaged even when use too many safe layer adjustments?
Steve - I'm not sure exactly what you're getting at. What do you mean by "are pixels damaged"?
Assuming you're talking about a TIFF or JPG, pixels (or really their values) simply are. They have particular values of RGB for color, or brightness for B&W. Photomatix, layer modifications - anything really - simply do some calculation and change on these values based on various factors, like original values, values of surrounding pixels, etc.
Obviously some thing move a pixel's values further than others, and with very different results when applied across all pixels.
Photomatix or other HDR is a bit of a unique case, in that the possible range of the values is actually changed in the intermediate step (to create the actual high dynamic range 32-bit image). But the pixels still have RGB or brightness values - although they can't be directly shown on the screen. To do that they need to be mapped back into the values that the software and screen can interpret, via the tone-mapping steps.
So at the end of the day, you wind up with pixel values that are mathematically calculated via 1) a mixture of the original pixel values for the multiple shot inputs that are then 2) mapped into a 32-bit space which can contain (mathematically) a massive dynamic range and 3) mapped again back down to a range that can be displayed, but with more apparent dynamic range than any 1 input image.
I'm not sure if that helps at all...maybe some more specific questions would help?
David,
It's my understanding that almost all image editing damages image pixels to one degree or another. Most damage is undetectable to the eye, but one can see it occur after heavy handed editing (esp. 8 bit files/JPEGs). That's why using non-destructive adjustment layers in PS is very strongly recommended.
Am I off base here?
I think you're right, John, but I'm still not certain what people mean by "damage" all the time, and I think sometimes they don't understand what they mean. At an individual color pixel level, the pixel is simply three numbers. Changing them doesn't really do anything but change their tone.
I think I know what people mean by damage sometimes, and usually its artifacts (pixels that jump strangely in value compared to surrounding one, or show unintended values that don't fit with those around them). This would usually be from compression schemes (JPG is one) that don't faithfully reproduce things like smooth gradients and fine detail. But it could occur from other tools/steps in processing images - like strange curves, or S/H tools. My point though, is that the image may look funky, but the image pixels have simply had their values shifted. Is that "damage"? I guess you could look at it that way. I just don't want anyone thinking that a pixel is anything more than it is.
And, as far as I'm concerned, the main reason to use non-destructive editing is that any changes can be undone or modified at any stage, so that changes that you think have a negative effect can easily be fixed. And you can always start at the actual camera capture if you want, which should be closer to "natural" or what your eyes see. It's at least what the camera saw, and hasn't had any functions performed on the pixel values, and I suppose is therefore less "damaged".
I couldn't agree more with David's explanation. The "damaging pixels" phrase gets thrown around quite often, but I have never understood what people mean by it. In the simplest terms, a pixel is a square of color (or shade of grey for b&w). No actual damage can be done to it. There are really only two things you can do to pixels... change their color value or remove them (cropping). Many adjustments can be made that negatively affect the way a smooth gradient appears (artifacts, banding) and the way detail and edges are displayed (halos, noise) but you can't actually damage a pixel. I assume people are referring to the artifacts, noise, banding, etc when they say "damage".
As far as adjustment layers go, I don't think they are less "damaging" or are "safer" than a regular "on-layer" adjustment. If you make a curves adjustment to an actual pixel containing layer, it will have the exact same result/affect as if you made the same adjustment on an adjustment layer. The only benefit to using adjustment layers is the ability to go back and change these at a later time. Somebody please let me know if I'm wrong about this.
As far as HDR vs. hand blending exposures, I've never had the chance to check out an HDR file myself, nor do I really understand how the software works, so I couldn't comment factually. I do see a lot of HDR images that have quite a bit of noise, both luminosity and color noise. I'd assume this is from the shadows being brought up so much... which could be user inflicted and not necessarily the HDR software's fault. However, I have seen quite a few guys getting great results from Photomatix's Exposure Fusion process. Again, I've never had the chance to look at an HDR file from Photomatix software, so I really don't know much here. I have seen one 24x36in print that was created with a 5d2 and HDR, it looked alright... fairly decent, but definitely not of the quality as the prints I get from my 5d2 and hand blending. But to give Photomatix the benefit of the doubt, it could have been some other processing work or poor printing technique that caused the mediocre print.
Two things: 1) isn't it true that converting a RAW capture and saving as a jpg actually discards data? Isn't that data then pixels that get thrown away? 2) I thought adjustment layers were "instructions" on how to make a pixel appear to us or to be reproduced (printed) rather than an alteration to the values, numbers as has been said here. So in this case, a curves adjustment to the pixel layer changes the numbers and an adjustment layer instead does not but does some math on the numbers. Is this not so?
Just thought of something else...isn't resampling adding new pixels so that even though the original pixels' values may not be changed, the overall image quality can be due to resampling approximations in the "new" pixels?
Thanks.
John,
RE: your first query, yes converting to jpg file format discards data--not in the RAW file--but in the one-off copy. Whether the discarded data actually matters is a topic for debate. It depends on the reproduction requirements. For web delivery, it doesn't really matter. A useful analogy would be converting music on CD to MP3 or AAC formats. Yes, a lot of musical data gets tossed, but if one is simply listening through speakers in a noisy automotive interior (and you ears are 4-5 decades old), then does it really matter?
1) isn't it true that converting a RAW capture and saving as a jpg actually discards data? Isn't that data then pixels that get thrown away?
Well there are actually two different things that could be referred to as data loss when going from RAW to JPG.
1) RAW data is actual sensor data in a bayer array pattern. To get to JPG (or TIFF, or even the image in your RAW converter), this data has to be demosaiced (meaning it goes from "single color", or brightness only, bayer array pixels, to full color, 3-value, pixels). This demosaicing could be referred to as a data loss, since you no longer have exactly the information the camera captured. But you do have the same amount of pixels, since each pixel's new value is calculated from it's captured single-color value plus information from those around it.
2) The other place is in applying the JPG compression algorithms, which are "lossy", because they do "throw away" data. But in this case the "data" being referred to is information about the exact values of pixels in the original image.
And it's only lost because the algorithm for JPG makes approximations that means when you save, close, and then re-open the image, there's no way for your software to get back to the original values. That's what "lossy" compression means. TIFF compression (ZIP or LZW) is not lossy, so your software can re-produce the exact values when the file is repeatedly saved and re-opened.
So pixels themselves are not lost. When you re-open a JPG, they simply have had their values changed in a way that doesn't necessarily reference their original values. (Hence the output image is the same resolution as the input image.) Those original values are the information that is "lost". It's damaging, because a by-product of losing reference to the original pixels is that you lose fine detail. If you save, close, and re-open a JPG image a bunch of times (probably like 10-20 depending on the image) you'll start to see a blurry, pixelated, poor looking image.
There is one other point that can be added to the issue of "pixel quality", aside from the mention of demosaicing and JPEG conversions. That's possible round-off errors in the math as you process an image. I've seen reference to this in some books and online articles, although I cannot provide a particular reference at the moment. (Sorry.) In such cases, I believe that you start reducing the accuracy of the original data the more layers and filters you apply in Photoshop. It can even show up in a raw converter like ACR, which is one of the arguments Andrey Tverdokhleb gave for using floating point arithmetic in his Raw Photo Processor (RPP).
Of course, whether you or I can actually see the differences is another matter. It's a bit like adjusting out-of-gamut colors using the ProPhoto RGB color space when your monitor cannot show the colors. Perhaps all this is for purists only. ;-)
Dave - There are lots of those kinds of errors, "behind-the-scenes" so to speak, that can occur. The examples you gave are real and can cause issues, but are very rare. And when they do occur, might not even be visible, which is why we don't have to worry about them. Thankfully!
Posted by Steve Sieren on May 18, 2010 at 12:57 AM
Sorry guys, I took off for the weekend. I was curious to what really happens to pixels in the photomatix way of blending. Where does the graininess come from?
What I mean by damage is loss of sharpness and noise or grain added to each pixel or group of pixels. Where does loss of sharpness happen in the blending process. If a gaussian blur is used as part of a blending process photomatix or hand blending are pixels losing their sharpness?
Well sharpness is not really a property of individual pixels (just little boxes of color after all), but is the difference between adjacent or nearby pixels where there are edges or detail. I don't know Photomatix well enough to pinpoint where any loss of sharpness could occur, but it's completely possible that the algorithms used could calculate the new pixel values in such a way that edges/detail get's smoothed out to some extent.
What you could also be seeing is the overall loss in contrast that can easily occur with an HDR shot. Lower contrast images will appear to your eye to be less sharp no matter what, since the edges become less well-defined. Since you're compressing a larger dynamic range into a smaller one, local areas will get squeezed into having less contrast. I find that adding extra sharpening and clarity (in ACR - or local contrast sharpening with USM in PS) can really make up for it and restore some "pop" while still maintaining the high dynamic range.
Do you have any specific before-and-after type shots?
Like David said, pixels can't actually loose sharpness... they are, and can never cease from being, squares of color. Think of it like those pictures people make using only small dots... kinda the same concept. Shapness is based on how quick the transition is from one different contrasting group of pixels to the next. For example your image you posted, the actual pixels will never become soft, but the transition from rock to sky can, as well as the edges from the little shadowy lines in the rock to more lit rock. If the edge from rock to sky transitions in a matter of 2-4 pixels it will be quite sharp, if it changes gradually from rock to sky in a matter of 8 pixels it will appear softer.
As for gaussian blur in the editing process, this can be used at a low opacity without sacrificing detail. Like in using a "glow" or "orton" technique... at a low opacity you will still have your sharpess, but the blur will ever so lightly bleed the lights into darks (and vice versa) on high contrast edges... softening the apparent transition of light without actually softening the sharpness. The bleeding is also done quite widely (again at a low opacity) so it doesn't have the same look as if the edge was actually soft from being out of focus.
I can't really comment on where the noise comes from in an hdr process, I have no experience with or knowledge of the program. I know some people really like it, and I've heard the new HDR pro (I think that's what it's called) in CS5 is working really well for some guys. But I'm with you, HDR seems to introduce unwanted noise and funky edges...
I wish I had 1/10 of the knowledge shared on these pages.
Thanks for taking the time to post and to allow others to read.
the fisch
Posted by Steve Sieren on May 21, 2010 at 10:45 PM
David, I do not have any files at the moment and I don't have a copy of photomatix at the time. The file was an old file when I did play around with photomatix. I have files that did work and some that didn't. Some files you couldn't even tell that were HDR. Amazing how that works out occasionly.
Ryan, a square is a square and that will not change but groups of pixels that make up lines or luminance edges get soft from what you saying. Photomatix blends flies together that aren't exactly the same even mounted perfectly on a tripod. Any of the stark luminance edges differ between each file and shift due to the different intervals of light reaching the sensor on seperate exposures. It must soften luminance edges because files many stops apart that do not have the exact information in specific parts of the file. It has to guess what is right and that is when the errors can happen bringing out halos to a highly varied amount.
Now here comes my opinion on adding a dreamy effect. I definitely know that orton or glow is no different from the old school bridal shots that have the ethereal effects going and you are removing sharpness which in portraiture is very normal. In Landscapes we pay big bucks to get sharpness throughout a scene . There are ways around the orton and glow effect to punch contrast and luminosity. The more that opacity slider moves and adds from a guassian blur layer the more it degrades the quality of what a lens captured so you are sacrificing detail with gaussian blur to a certain degree when adding glow and it will highly vary from what you see on screen versus in print. If you use this on all of your images it would lessen the impact for your prospective buyers because everything is the same and nothing varies especially if your competitors do the same exact thing.
From what most people might not know a halo is normal even for single shot images. Take a look at a RAW capture of a seascape and check the horizon. There will always be a stripe there that is one pixel high where sky and water connect.
Noise in HDR imagery shouldn't be there since all detail is exposed for correctly but that is another story for another time.
Here is a microscopic example where at the bottom portion it was block of solid white and solid blue. Anyone can copy the file and open it in photoshop but they will have to zoom in since it's only 8x24 pixels (all still square). It is a luminance edge after all of the non destructive layer adjustments. You can see what is was like from the beginning at the bottom. The top has changed, solid white has moved from 255 on the color palette to 247 and the blue has changed even more dramatically on the color palette. A pixel is always square but pixels as a group do change and does the quality of the group change? I'm still wondering... . .
Posted by Steve Sieren on May 21, 2010 at 11:00 PM
Here is a comparison, I won't say which is which but one is an untouched raw capture and the other is an HDR with a little bit of glamor glow added to cover up some of the noise and graininess. One has more dynamic range with the hot spots and it could still use more of a punch but it's a small piece of scene zoomed it a 100%.
Discussion
16 repliesI'm not trying to bite anyone's head off. I'm just looking for opinions and facts. What actually happens to each individual pixel when you do use photomatix or other blending techniques? Are pixels damaged even when use too many safe layer adjustments?
Assuming you're talking about a TIFF or JPG, pixels (or really their values) simply are. They have particular values of RGB for color, or brightness for B&W. Photomatix, layer modifications - anything really - simply do some calculation and change on these values based on various factors, like original values, values of surrounding pixels, etc.
Obviously some thing move a pixel's values further than others, and with very different results when applied across all pixels.
Photomatix or other HDR is a bit of a unique case, in that the possible range of the values is actually changed in the intermediate step (to create the actual high dynamic range 32-bit image). But the pixels still have RGB or brightness values - although they can't be directly shown on the screen. To do that they need to be mapped back into the values that the software and screen can interpret, via the tone-mapping steps.
So at the end of the day, you wind up with pixel values that are mathematically calculated via 1) a mixture of the original pixel values for the multiple shot inputs that are then 2) mapped into a 32-bit space which can contain (mathematically) a massive dynamic range and 3) mapped again back down to a range that can be displayed, but with more apparent dynamic range than any 1 input image.
I'm not sure if that helps at all...maybe some more specific questions would help?
It's my understanding that almost all image editing damages image pixels to one degree or another. Most damage is undetectable to the eye, but one can see it occur after heavy handed editing (esp. 8 bit files/JPEGs). That's why using non-destructive adjustment layers in PS is very strongly recommended.
Am I off base here?
I think I know what people mean by damage sometimes, and usually its artifacts (pixels that jump strangely in value compared to surrounding one, or show unintended values that don't fit with those around them). This would usually be from compression schemes (JPG is one) that don't faithfully reproduce things like smooth gradients and fine detail. But it could occur from other tools/steps in processing images - like strange curves, or S/H tools. My point though, is that the image may look funky, but the image pixels have simply had their values shifted. Is that "damage"? I guess you could look at it that way. I just don't want anyone thinking that a pixel is anything more than it is.
And, as far as I'm concerned, the main reason to use non-destructive editing is that any changes can be undone or modified at any stage, so that changes that you think have a negative effect can easily be fixed. And you can always start at the actual camera capture if you want, which should be closer to "natural" or what your eyes see. It's at least what the camera saw, and hasn't had any functions performed on the pixel values, and I suppose is therefore less "damaged".
As far as adjustment layers go, I don't think they are less "damaging" or are "safer" than a regular "on-layer" adjustment. If you make a curves adjustment to an actual pixel containing layer, it will have the exact same result/affect as if you made the same adjustment on an adjustment layer. The only benefit to using adjustment layers is the ability to go back and change these at a later time. Somebody please let me know if I'm wrong about this.
As far as HDR vs. hand blending exposures, I've never had the chance to check out an HDR file myself, nor do I really understand how the software works, so I couldn't comment factually. I do see a lot of HDR images that have quite a bit of noise, both luminosity and color noise. I'd assume this is from the shadows being brought up so much... which could be user inflicted and not necessarily the HDR software's fault. However, I have seen quite a few guys getting great results from Photomatix's Exposure Fusion process. Again, I've never had the chance to look at an HDR file from Photomatix software, so I really don't know much here. I have seen one 24x36in print that was created with a 5d2 and HDR, it looked alright... fairly decent, but definitely not of the quality as the prints I get from my 5d2 and hand blending. But to give Photomatix the benefit of the doubt, it could have been some other processing work or poor printing technique that caused the mediocre print.
Just my two cents, take it for what it's worth.
Just thought of something else...isn't resampling adding new pixels so that even though the original pixels' values may not be changed, the overall image quality can be due to resampling approximations in the "new" pixels?
Thanks.
RE: your first query, yes converting to jpg file format discards data--not in the RAW file--but in the one-off copy. Whether the discarded data actually matters is a topic for debate. It depends on the reproduction requirements. For web delivery, it doesn't really matter. A useful analogy would be converting music on CD to MP3 or AAC formats. Yes, a lot of musical data gets tossed, but if one is simply listening through speakers in a noisy automotive interior (and you ears are 4-5 decades old), then does it really matter?
ME
Well there are actually two different things that could be referred to as data loss when going from RAW to JPG.
1) RAW data is actual sensor data in a bayer array pattern. To get to JPG (or TIFF, or even the image in your RAW converter), this data has to be demosaiced (meaning it goes from "single color", or brightness only, bayer array pixels, to full color, 3-value, pixels). This demosaicing could be referred to as a data loss, since you no longer have exactly the information the camera captured. But you do have the same amount of pixels, since each pixel's new value is calculated from it's captured single-color value plus information from those around it.
2) The other place is in applying the JPG compression algorithms, which are "lossy", because they do "throw away" data. But in this case the "data" being referred to is information about the exact values of pixels in the original image.
And it's only lost because the algorithm for JPG makes approximations that means when you save, close, and then re-open the image, there's no way for your software to get back to the original values. That's what "lossy" compression means. TIFF compression (ZIP or LZW) is not lossy, so your software can re-produce the exact values when the file is repeatedly saved and re-opened.
So pixels themselves are not lost. When you re-open a JPG, they simply have had their values changed in a way that doesn't necessarily reference their original values. (Hence the output image is the same resolution as the input image.) Those original values are the information that is "lost". It's damaging, because a by-product of losing reference to the original pixels is that you lose fine detail. If you save, close, and re-open a JPG image a bunch of times (probably like 10-20 depending on the image) you'll start to see a blurry, pixelated, poor looking image.
Of course, whether you or I can actually see the differences is another matter. It's a bit like adjusting out-of-gamut colors using the ProPhoto RGB color space when your monitor cannot show the colors. Perhaps all this is for purists only. ;-)
Dave
What I mean by damage is loss of sharpness and noise or grain added to each pixel or group of pixels. Where does loss of sharpness happen in the blending process. If a gaussian blur is used as part of a blending process photomatix or hand blending are pixels losing their sharpness?
What you could also be seeing is the overall loss in contrast that can easily occur with an HDR shot. Lower contrast images will appear to your eye to be less sharp no matter what, since the edges become less well-defined. Since you're compressing a larger dynamic range into a smaller one, local areas will get squeezed into having less contrast. I find that adding extra sharpening and clarity (in ACR - or local contrast sharpening with USM in PS) can really make up for it and restore some "pop" while still maintaining the high dynamic range.
Do you have any specific before-and-after type shots?
As for gaussian blur in the editing process, this can be used at a low opacity without sacrificing detail. Like in using a "glow" or "orton" technique... at a low opacity you will still have your sharpess, but the blur will ever so lightly bleed the lights into darks (and vice versa) on high contrast edges... softening the apparent transition of light without actually softening the sharpness. The bleeding is also done quite widely (again at a low opacity) so it doesn't have the same look as if the edge was actually soft from being out of focus.
I can't really comment on where the noise comes from in an hdr process, I have no experience with or knowledge of the program. I know some people really like it, and I've heard the new HDR pro (I think that's what it's called) in CS5 is working really well for some guys. But I'm with you, HDR seems to introduce unwanted noise and funky edges...
Thanks for taking the time to post and to allow others to read.
the fisch
Ryan, a square is a square and that will not change but groups of pixels that make up lines or luminance edges get soft from what you saying. Photomatix blends flies together that aren't exactly the same even mounted perfectly on a tripod. Any of the stark luminance edges differ between each file and shift due to the different intervals of light reaching the sensor on seperate exposures. It must soften luminance edges because files many stops apart that do not have the exact information in specific parts of the file. It has to guess what is right and that is when the errors can happen bringing out halos to a highly varied amount.
Now here comes my opinion on adding a dreamy effect. I definitely know that orton or glow is no different from the old school bridal shots that have the ethereal effects going and you are removing sharpness which in portraiture is very normal. In Landscapes we pay big bucks to get sharpness throughout a scene . There are ways around the orton and glow effect to punch contrast and luminosity. The more that opacity slider moves and adds from a guassian blur layer the more it degrades the quality of what a lens captured so you are sacrificing detail with gaussian blur to a certain degree when adding glow and it will highly vary from what you see on screen versus in print. If you use this on all of your images it would lessen the impact for your prospective buyers because everything is the same and nothing varies especially if your competitors do the same exact thing.
From what most people might not know a halo is normal even for single shot images. Take a look at a RAW capture of a seascape and check the horizon. There will always be a stripe there that is one pixel high where sky and water connect.
Noise in HDR imagery shouldn't be there since all detail is exposed for correctly but that is another story for another time.
Here is a microscopic example where at the bottom portion it was block of solid white and solid blue. Anyone can copy the file and open it in photoshop but they will have to zoom in since it's only 8x24 pixels (all still square