For years I've been looking a my blog and telling myself I should do something about the format that forced me to post ridiculously small images. After this was a blog about photography. But I never did anything except add a line saying 'click and thou shall see large.'
Part was 'procrastination thy name is scribble.' Part was an ancient fear from olden days that I might end up having to edit some HTML code. Been there. Done that. And folks I do know how to mess up HTMl code big time.Then, in the spirit of spring cleaning, I poked around a bit in Blogger and saw how simple it was. A new template, a few slider adjustments and done.
Might even do a post or two with just photos and no theoretical stuff. Now that is a wild idea.
Monday, April 4, 2011
Tuesday, March 29, 2011
D7000 noise- part 2
Quick question. Which patch has the least amount of noise. The one with the line profile under the image? Or the one with the line profile beside the image?
Actually nether. It's the same patch. In the bottom graph I measured 125 pixels. In the side graph I dragged the line out so I measured 1200 pixels. Because ten times more noise wiggles were compress together the line is darker and looks noisier. This complicates things especially when comparing cameras with differing numbers of pixels.
From now on graphs are for info only. I'll be posting result boxes showing the rms mean value (signal) and sdev (standard deviation or noise) Using these numbers I'll calculate the S/N numbers for any comparisons.
Yesterday I ate a few crow feathers in the D7000 thread. I'd blindly taken imaging-resource's word that they had measure the lighting for the D700 ISO6400 image. One look at the EXIF info and it was obvious that instead of 11 lumens, the lighting you find on a well lite street at night--the lighting used was more like 30 lumens.
My only excuse for such a stupid mistake is up to now I have been using my blog as an online lab notebook. So I was free to changed my mine latter if I goofed. After all nobody read my theoretical posts. Until this one caught peoples' attention. So my apologies for any confusion and on to what really is happening.
A way to correct for the lighting differences was to search out for areas in the images that had the same signal and then compare the noise. When I did it for the highlights, the black circled areas, it was definitely Oops. The D700 beat out the D7000 by coincidentally 1.44--the magic noise number for a pixel twice bigger than the D7000's pixel.
Things become more interesting when I did the same thing for the shadows. I chose the second dark patch for the D700 (blue) and an area on the frame of the Dave Box for the D7000 (red). The result box says it all. Down in the shadows the S/N was identical.
I suddenly had some interesting ideas on what was going on, but since I was dealing with one set of suspect data, I certainly didn't have a convincing case. Even in my own mind.
Then d.brodsky posted a link to his images of the same scene, one taken with a D7000 and one with a D700. He provided a second data set.
(Click on the clips for full size and readable images)
The EXIF info says the images were taken with two different lens so the exposures weren't exactly identical but I did find areas where the signal values were close. The result boxes show that with real world images the noise reduction in the two cameras can smooth the noise out to were the signal to noise is nearly identical.
So my current thinking that Nikon's noise reductions routines are sophisticated. Besides the obvious settings-- off, low, normal and high-- the algorithm looks at the S/N in the various areas of the image. If it is in a highlight were the S/N is high it uses little or no noise reduction. If it is the shadows it uses a lot.
So if you compare images by looking at the bright areas where the noise is easy to see, the pixel size difference is important. But if you are digging detail out of the shadows the two camera may give the same results.
My plans are to uses my D7000 to figure out what is going on. But it might be a while before I post again since the computer I'm working with is going back to the shop for warranty repairs. Who knows how long that will take.
And finally a special thanks to d.brodsky for sharing his images.
Actually nether. It's the same patch. In the bottom graph I measured 125 pixels. In the side graph I dragged the line out so I measured 1200 pixels. Because ten times more noise wiggles were compress together the line is darker and looks noisier. This complicates things especially when comparing cameras with differing numbers of pixels.
From now on graphs are for info only. I'll be posting result boxes showing the rms mean value (signal) and sdev (standard deviation or noise) Using these numbers I'll calculate the S/N numbers for any comparisons.
Yesterday I ate a few crow feathers in the D7000 thread. I'd blindly taken imaging-resource's word that they had measure the lighting for the D700 ISO6400 image. One look at the EXIF info and it was obvious that instead of 11 lumens, the lighting you find on a well lite street at night--the lighting used was more like 30 lumens.
My only excuse for such a stupid mistake is up to now I have been using my blog as an online lab notebook. So I was free to changed my mine latter if I goofed. After all nobody read my theoretical posts. Until this one caught peoples' attention. So my apologies for any confusion and on to what really is happening.
A way to correct for the lighting differences was to search out for areas in the images that had the same signal and then compare the noise. When I did it for the highlights, the black circled areas, it was definitely Oops. The D700 beat out the D7000 by coincidentally 1.44--the magic noise number for a pixel twice bigger than the D7000's pixel.
Things become more interesting when I did the same thing for the shadows. I chose the second dark patch for the D700 (blue) and an area on the frame of the Dave Box for the D7000 (red). The result box says it all. Down in the shadows the S/N was identical.
I suddenly had some interesting ideas on what was going on, but since I was dealing with one set of suspect data, I certainly didn't have a convincing case. Even in my own mind.
Then d.brodsky posted a link to his images of the same scene, one taken with a D7000 and one with a D700. He provided a second data set.
(Click on the clips for full size and readable images)
The EXIF info says the images were taken with two different lens so the exposures weren't exactly identical but I did find areas where the signal values were close. The result boxes show that with real world images the noise reduction in the two cameras can smooth the noise out to were the signal to noise is nearly identical.
So my current thinking that Nikon's noise reductions routines are sophisticated. Besides the obvious settings-- off, low, normal and high-- the algorithm looks at the S/N in the various areas of the image. If it is in a highlight were the S/N is high it uses little or no noise reduction. If it is the shadows it uses a lot.
So if you compare images by looking at the bright areas where the noise is easy to see, the pixel size difference is important. But if you are digging detail out of the shadows the two camera may give the same results.
My plans are to uses my D7000 to figure out what is going on. But it might be a while before I post again since the computer I'm working with is going back to the shop for warranty repairs. Who knows how long that will take.
And finally a special thanks to d.brodsky for sharing his images.
Monday, March 28, 2011
D7000-Best low light camera ever?
Edit Learned a few things since I posted this. For the current thinking also see the next post: D7000 noise part 2
In a recent D7000 club post a low light handhold photographer wished he could afford the $2700 for a D700 body so he could take even lower light handhold photos. Since taking low light handheld photos is one of my things I became curious about how much better the full frame D700 is vs the cropped D7000. So I did some Internet research.
Everybody seems to go to dpreview nowadays but when it comes to comparing two camera's low light performance no review site beats imaging-resource. Why? Their Dave Box. They have been posting full size jpg images of that well designed target under pretty much the same lighting since 1995. Combine it with ImageJ-free NIH image analysis software- and you have everything you need to separate the good low noise cameras from the bad. And with hardly any work.
An A/B comparison of the d7000 vs the D700 Click on the image to see the full size screen clip.
What does the clip show? The smaller graph is the D7000. The larger is the D700 which I stretched out a little because it's actual ISO6400 is about 12% less than the D7000's 6400 ISO. The noise profiles were taken on the five gray patches and show the RGB signal and noise. (See my Nov 2009 post 'Fun with ImageJ' for a procedure on how I did the measurement.)
Bottom line--save your money. When it comes to ISO 6400 jpgs straight out of the camera with the default ISO noise reduction set to normal there isn't a nits bit of difference in the noise.
And if you want to really waste money go for a $9000 D3x. With that older design the 6400 ISO signal to noise is worse.
How do I defend my heresy--that big sensors aren't the greatest of the great. With the jpgs that we print up or post the S/N is a combination of many factors. But with modern DSLR cameras the wavelet noise reduction in the firmware trumps them all. The myth that big sensors are always less noisy is a hold over from the days before decent noise reduction routines. Even in the old day, the noise difference wasn't all that much. Double the sensor size with a big jump in price and you only end up with 40% less photon shot noise. (See my two Nov 2009 posts on noise for a fuller explanation.)
This is assuming the noise reduction is working. When you look at the S/N in the D3x white patch you might think that's super great except for the little spike at the beginning. Ain't true. When the noise reduction when on stage to do its strip-out-the-noise dance it got carried away. Went totally nude. What you are seeing is flat line with no detail.
Which brings up the fine art of marketing a camera. The marketing folks try to send a sample camera to a review site as soon as possible so great reviews appear as the camera hits the stores. And if the camera they send isn't quite ready for prime time? Who cares. Unless someone does measurements that patch looks white and OK.
I assume the firmware was fixed before it hit the stores. Us ordinary folks might be fooled but a pro is going to notice and not put up with that sort of performance.
Here are the details. The images I downloaded are found in the sample section of the camera review. The lighting is supposed to be 1 foot candle or 11 lumens--they messed up a little in the D700 review. Eleven lumens is what you find in a well lit city street at night. They also post a full series of shots where they cut the lighting back a stop at a time to show what you will see under worse conditions.
How accurate are the Imaging-resource images? Here are the measurements on a photo of a young rock hound I took a few days after I bought my D7000. ISO 6400 and NR normal. I've circled the areas where I took the noise profiles. If anything my camera maybe be slightly less noisy than the one used in the Imaging-resource review.
Edit- I forgot to explain the result box in the top photo. If you look at the S/N in the D7000's second darkest patch you might say it's worse than the D700's S/N. But when you calculate the rms S/N (mean/sdev) the D7000's S/N is 20 and the D700's S/N is 17. A minor quirk of the measurement and not significant either way. After all we are dealing with random noise.
In a recent D7000 club post a low light handhold photographer wished he could afford the $2700 for a D700 body so he could take even lower light handhold photos. Since taking low light handheld photos is one of my things I became curious about how much better the full frame D700 is vs the cropped D7000. So I did some Internet research.
Everybody seems to go to dpreview nowadays but when it comes to comparing two camera's low light performance no review site beats imaging-resource. Why? Their Dave Box. They have been posting full size jpg images of that well designed target under pretty much the same lighting since 1995. Combine it with ImageJ-free NIH image analysis software- and you have everything you need to separate the good low noise cameras from the bad. And with hardly any work.
An A/B comparison of the d7000 vs the D700 Click on the image to see the full size screen clip.
What does the clip show? The smaller graph is the D7000. The larger is the D700 which I stretched out a little because it's actual ISO6400 is about 12% less than the D7000's 6400 ISO. The noise profiles were taken on the five gray patches and show the RGB signal and noise. (See my Nov 2009 post 'Fun with ImageJ' for a procedure on how I did the measurement.)
Bottom line--save your money. When it comes to ISO 6400 jpgs straight out of the camera with the default ISO noise reduction set to normal there isn't a nits bit of difference in the noise.
And if you want to really waste money go for a $9000 D3x. With that older design the 6400 ISO signal to noise is worse.
How do I defend my heresy--that big sensors aren't the greatest of the great. With the jpgs that we print up or post the S/N is a combination of many factors. But with modern DSLR cameras the wavelet noise reduction in the firmware trumps them all. The myth that big sensors are always less noisy is a hold over from the days before decent noise reduction routines. Even in the old day, the noise difference wasn't all that much. Double the sensor size with a big jump in price and you only end up with 40% less photon shot noise. (See my two Nov 2009 posts on noise for a fuller explanation.)
This is assuming the noise reduction is working. When you look at the S/N in the D3x white patch you might think that's super great except for the little spike at the beginning. Ain't true. When the noise reduction when on stage to do its strip-out-the-noise dance it got carried away. Went totally nude. What you are seeing is flat line with no detail.
Which brings up the fine art of marketing a camera. The marketing folks try to send a sample camera to a review site as soon as possible so great reviews appear as the camera hits the stores. And if the camera they send isn't quite ready for prime time? Who cares. Unless someone does measurements that patch looks white and OK.
I assume the firmware was fixed before it hit the stores. Us ordinary folks might be fooled but a pro is going to notice and not put up with that sort of performance.
Here are the details. The images I downloaded are found in the sample section of the camera review. The lighting is supposed to be 1 foot candle or 11 lumens--they messed up a little in the D700 review. Eleven lumens is what you find in a well lit city street at night. They also post a full series of shots where they cut the lighting back a stop at a time to show what you will see under worse conditions.
How accurate are the Imaging-resource images? Here are the measurements on a photo of a young rock hound I took a few days after I bought my D7000. ISO 6400 and NR normal. I've circled the areas where I took the noise profiles. If anything my camera maybe be slightly less noisy than the one used in the Imaging-resource review.
Edit- I forgot to explain the result box in the top photo. If you look at the S/N in the D7000's second darkest patch you might say it's worse than the D700's S/N. But when you calculate the rms S/N (mean/sdev) the D7000's S/N is 20 and the D700's S/N is 17. A minor quirk of the measurement and not significant either way. After all we are dealing with random noise.
Subscribe to:
Posts (Atom)