Achieving FF IQ from Nikon 1 system

Horshack

Forum Pro
Messages
11,231
Solutions
28
Reaction score
12,594
Location
US
I recently acquired a J4 and a preliminary workflow for producing full-frame equivalent IQ (in terms of noise). There are two main aspects:

ETTR (Expose to the Right)

Exposing the scene so that the brightest highlights wanted/needed are near the rightmost edge of the raw data/histogram. Unfortunately the J4 doesn't provide an RGB histogram and its luminance histogram is imprecise because it's based on a JPEG tone curve (even when shooting raw). I haven't yet tried UniWB or altering the picture control to produce a more reliable histogram. For now I just calibrated the raw levels vs the J4's middle-gray exposure for my most-often shooting situation of landscapes. I've found that middle-gray + 3EV will yield a relatively-safe ETTR exposure, provided the sky is the brightest element of the scene.

When shooting I have the camera's metering set to spot meter and the shooting mode to Creative-Manual. I meter the sky to bring the exposure scale to "0", then adjust the exposure to 3EV brighter than that. This is a little clumsy for repeated shooting but if you're taking multiple shots in the same lighting situation you only need to perform the metering once. A more convenient alternative would be to use Creative-Aperture Priority and keep the exposure compensation pegged to +3EV - that way the camera would always meter +3EV when you point to the sky. Unfortunately the J4 has a bug where exposure compensation to near saturation levels causes the meter to over-exposure by up to 2EV. Not sure if other 1-series bodies have this issue.

Image Stacking for noise reduction

In addition to ETTR I also shoot 20-shot burst for each photograph (you can use a 40-shot on the V3 for even better noise reduction). For this I set the mode dial to "Best Moment Capture" and the capture mode to "Active Selection". For the "Active Selection" option I use the "On release, record preceding frames". That way I can hold the shutter half-pressed and have all 20 frames captured when I release the shutter, producing less motion jerking of the camera. I have the "Capture 20 frames over" parameter set to 1/3 second instead of 1 second - you want to capture all 20 images as quickly as possible because even when using a fast shutter speed the camera is only able to read the sensor at 1/60. I set the "Frames saved(default)" to ALL so the camera defaults to store all 20 images of the burst.

Post-Processing

All 20 images are dropped into Photoshop ACR. I set the WB, sharpening and general adjustments on the first image, then synchronize the other 19 images to the first. I then click "done" and use File -> Scripts -> Load Files Into Stack. I click both "Attempt to Automatically Align Source Images" (since I shoot hand-held there are a few pixels of shift between images sometimes) and "Create Smart Object after Loading Layers". The script takes a few minutes to run. When done, I then go to Layer -> Smart Objects -> Stack Mode -> Median. That takes about 30 seconds. When done you can then crop the image to crop off any blank areas from any pixel shift between the 20 images - I intentionally do this after the Stack Mode because that fills in some of the missing pixels first. Note that a 20-image stack with median averaging reduces noise by 3.5 stops. A 40-image stack (V3) would reduce noise by 5.3 stops. AE bracketing/blending can achieve the same level of noise reduction and only require a few frames but the J-series doesn't support bracketing - plus the stacking technique is more suitable for High ISO hand-held situations as well.

Results

Here are the results of a sample scene. I used a J4 with the 10-30mm PD kit lens and for comparison, a D750 with a Tamron 24-70 VC. I set the aperture, shutter speed, and ISO the same for both bodies. Note that the ETTR on the D750 caused some highlight loss but I wanted to keep the exposure the same so that the noise comparison is fair to the D750. All had the ACR shadow slider increased to max +100 (for the bushes in the shade, simulating a High-DR scene), and all were sharpened in ACR with 35/35 and two passes of USM in PS (30/30/0 and then 300/0.5/0)

[IMG width="400px" alt="100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack."]http://horshack.smugmug.com/photos/i-VFkQHpH/0/O/i-VFkQHpH.jpg[/IMG]
100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack.

And here are full-sized images of the image, downsampled to 8MP:

J4 Matrix-Metered (single exposure)

J4 ETTR (single exposure)

J4 Matrix-Metered (20-image stack)

J4 ETTR (20-image stack)

D750 Matrix-Metered (single exposure)

D750 ETTR (single exposure)
 
Last edited:
Fantastic.

Thanks so much for sharing your research and the detailed description!
 
Fantastic.

Thanks so much for sharing your research and the detailed description!
The stacking method doesn't work well with things that move! I wish it did!
 
Last edited:
This one with the framing better matched and more shadow areas to compare. All three images below are ETTR and the shadows raised in post-processing.

[IMG width="400px" alt="100% crops from 8MP downsample. Click "original size" to view the crops at full-size."]http://horshack.smugmug.com/photos/i-HbJf8Mv/0/O/i-HbJf8Mv.jpg[/IMG]
100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)
...but all this is just a dream. The single D750 image is considerably better than the 20 exposure stack. There is no way to get around the sensor size advantage regarding DR. It's not working in reality, not even in your examples. This is clearly visible even in the tiny post stamp size image above, and even more clearly at "Original" size.
 
Last edited:
I recently acquired a J4 and a preliminary workflow for producing full-frame equivalent IQ (in terms of noise). There are two main aspects:

ETTR (Expose to the Right)

Exposing the scene so that the brightest highlights wanted/needed are near the rightmost edge of the raw data/histogram. Unfortunately the J4 doesn't provide an RGB histogram and its luminance histogram is imprecise because it's based on a JPEG tone curve (even when shooting raw). I haven't yet tried UniWB or altering the picture control to produce a more reliable histogram. For now I just calibrated the raw levels vs the J4's middle-gray exposure for my most-often shooting situation of landscapes. I've found that middle-gray + 3EV will yield a relatively-safe ETTR exposure, provided the sky is the brightest element of the scene.

When shooting I have the camera's metering set to spot meter and the shooting mode to Creative-Manual. I meter the sky to bring the exposure scale to "0", then adjust the exposure to 3EV brighter than that. This is a little clumsy for repeated shooting but if you're taking multiple shots in the same lighting situation you only need to perform the metering once. A more convenient alternative would be to use Creative-Aperture Priority and keep the exposure compensation pegged to +3EV - that way the camera would always meter +3EV when you point to the sky. Unfortunately the J4 has a bug where exposure compensation to near saturation levels causes the meter to over-exposure by up to 2EV. Not sure if other 1-series bodies have this issue.

Image Stacking for noise reduction

In addition to ETTR I also shoot 20-shot burst for each photograph (you can use a 40-shot on the V3 for even better noise reduction). For this I set the mode dial to "Best Moment Capture" and the capture mode to "Active Selection". For the "Active Selection" option I use the "On release, record preceding frames". That way I can hold the shutter half-pressed and have all 20 frames captured when I release the shutter, producing less motion jerking of the camera. I have the "Capture 20 frames over" parameter set to 1/3 second instead of 1 second - you want to capture all 20 images as quickly as possible because even when using a fast shutter speed the camera is only able to read the sensor at 1/60. I set the "Frames saved(default)" to ALL so the camera defaults to store all 20 images of the burst.

Post-Processing

All 20 images are dropped into Photoshop ACR. I set the WB, sharpening and general adjustments on the first image, then synchronize the other 19 images to the first. I then click "done" and use File -> Scripts -> Load Files Into Stack. I click both "Attempt to Automatically Align Source Images" (since I shoot hand-held there are a few pixels of shift between images sometimes) and "Create Smart Object after Loading Layers". The script takes a few minutes to run. When done, I then go to Layer -> Smart Objects -> Stack Mode -> Median. That takes about 30 seconds. When done you can then crop the image to crop off any blank areas from any pixel shift between the 20 images - I intentionally do this after the Stack Mode because that fills in some of the missing pixels first. Note that a 20-image stack with median averaging reduces noise by 3.5 stops. A 40-image stack (V3) would reduce noise by 5.3 stops. AE bracketing/blending can achieve the same level of noise reduction and only require a few frames but the J-series doesn't support bracketing - plus the stacking technique is more suitable for High ISO hand-held situations as well.

Results

Here are the results of a sample scene. I used a J4 with the 10-30mm PD kit lens and for comparison, a D750 with a Tamron 24-70 VC. I set the aperture, shutter speed, and ISO the same for both bodies. Note that the ETTR on the D750 caused some highlight loss but I wanted to keep the exposure the same so that the noise comparison is fair to the D750. All had the ACR shadow slider increased to max +100 (for the bushes in the shade, simulating a High-DR scene), and all were sharpened in ACR with 35/35 and two passes of USM in PS (30/30/0 and then 300/0.5/0)

[IMG width="400px" alt="100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack."]http://horshack.smugmug.com/photos/i-VFkQHpH/0/O/i-VFkQHpH.jpg[/IMG]
100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack.

And here are full-sized images of the image, downsampled to 8MP:

J4 Matrix-Metered (single exposure)
Is it normal for the J4 to be so noisy at ISO160 and in such good light and ideal conditions? Anyway, this image is overexposed.
Those two D750 images are Spot metered, NOT Matrix. Both are severely overexposed.
 
Last edited:
This one with the framing better matched and more shadow areas to compare. All three images below are ETTR and the shadows raised in post-processing.

[IMG width="400px" alt="100% crops from 8MP downsample. Click "original size" to view the crops at full-size."]http://horshack.smugmug.com/photos/i-HbJf8Mv/0/O/i-HbJf8Mv.jpg[/IMG]
100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)
...but all this is just a dream. The single D750 image is considerably better than the 20 exposure stack. There is no way to get around the sensor size advantage regarding DR. It's not working in reality, not even in your examples. This is clearly visible even in the tiny post stamp size image above, and even more clearly at "Original" size.
Stacking reduces image noise by the sqrt(# iamges), so a 20-image stack reduces noise by 3.5EV across the entire tonal range. DxO measures the D750 as having a 2.8EV SNR advtange (midtone noise) and 3.8EV DR advantage over the J4. After stacking 20 images the J4 has a 2/3EV SNR advantage over the D750 and is within 0.3EV of the D750 for DR, since engineering DR is increased by 3.8EV on the shadow end; photographic DR is increased even more on the J4 relative to the D750. Both the math and the images support the conclusion that the J4 matches FF noise performance when stacking images.
 
Last edited:
I recently acquired a J4 and a preliminary workflow for producing full-frame equivalent IQ (in terms of noise). There are two main aspects:

ETTR (Expose to the Right)

Exposing the scene so that the brightest highlights wanted/needed are near the rightmost edge of the raw data/histogram. Unfortunately the J4 doesn't provide an RGB histogram and its luminance histogram is imprecise because it's based on a JPEG tone curve (even when shooting raw). I haven't yet tried UniWB or altering the picture control to produce a more reliable histogram. For now I just calibrated the raw levels vs the J4's middle-gray exposure for my most-often shooting situation of landscapes. I've found that middle-gray + 3EV will yield a relatively-safe ETTR exposure, provided the sky is the brightest element of the scene.

When shooting I have the camera's metering set to spot meter and the shooting mode to Creative-Manual. I meter the sky to bring the exposure scale to "0", then adjust the exposure to 3EV brighter than that. This is a little clumsy for repeated shooting but if you're taking multiple shots in the same lighting situation you only need to perform the metering once. A more convenient alternative would be to use Creative-Aperture Priority and keep the exposure compensation pegged to +3EV - that way the camera would always meter +3EV when you point to the sky. Unfortunately the J4 has a bug where exposure compensation to near saturation levels causes the meter to over-exposure by up to 2EV. Not sure if other 1-series bodies have this issue.

Image Stacking for noise reduction

In addition to ETTR I also shoot 20-shot burst for each photograph (you can use a 40-shot on the V3 for even better noise reduction). For this I set the mode dial to "Best Moment Capture" and the capture mode to "Active Selection". For the "Active Selection" option I use the "On release, record preceding frames". That way I can hold the shutter half-pressed and have all 20 frames captured when I release the shutter, producing less motion jerking of the camera. I have the "Capture 20 frames over" parameter set to 1/3 second instead of 1 second - you want to capture all 20 images as quickly as possible because even when using a fast shutter speed the camera is only able to read the sensor at 1/60. I set the "Frames saved(default)" to ALL so the camera defaults to store all 20 images of the burst.

Post-Processing

All 20 images are dropped into Photoshop ACR. I set the WB, sharpening and general adjustments on the first image, then synchronize the other 19 images to the first. I then click "done" and use File -> Scripts -> Load Files Into Stack. I click both "Attempt to Automatically Align Source Images" (since I shoot hand-held there are a few pixels of shift between images sometimes) and "Create Smart Object after Loading Layers". The script takes a few minutes to run. When done, I then go to Layer -> Smart Objects -> Stack Mode -> Median. That takes about 30 seconds. When done you can then crop the image to crop off any blank areas from any pixel shift between the 20 images - I intentionally do this after the Stack Mode because that fills in some of the missing pixels first. Note that a 20-image stack with median averaging reduces noise by 3.5 stops. A 40-image stack (V3) would reduce noise by 5.3 stops. AE bracketing/blending can achieve the same level of noise reduction and only require a few frames but the J-series doesn't support bracketing - plus the stacking technique is more suitable for High ISO hand-held situations as well.

Results

Here are the results of a sample scene. I used a J4 with the 10-30mm PD kit lens and for comparison, a D750 with a Tamron 24-70 VC. I set the aperture, shutter speed, and ISO the same for both bodies. Note that the ETTR on the D750 caused some highlight loss but I wanted to keep the exposure the same so that the noise comparison is fair to the D750. All had the ACR shadow slider increased to max +100 (for the bushes in the shade, simulating a High-DR scene), and all were sharpened in ACR with 35/35 and two passes of USM in PS (30/30/0 and then 300/0.5/0)

[IMG width="400px" alt="100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack."]http://horshack.smugmug.com/photos/i-VFkQHpH/0/O/i-VFkQHpH.jpg[/IMG]
100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack.

And here are full-sized images of the image, downsampled to 8MP:

J4 Matrix-Metered (single exposure)
Is it normal for the J4 to be so noisy at ISO160 and in such good light and ideal conditions? Anyway, this image is overexposed.
Those two D750 images are Spot metered, NOT Matrix. Both are severely overexposed.
The exposure selected on the D750 was based on a matrix-metered reading which was identical between both cameras; the D750 happened to be in spot meter mode when I took the photograph and thus that is what's recorded in EXIF. The J4 is not clipped in the raw data so it is not overexposed; the D750 is clipped per my description to match the exposure of the J4. That was done so that the J4 wasn't given an unfair noise advantage. In the second sample I posted today I dialed back the ETTR on the cameras (to 2.3EV) to keep both within raw clipping.
 
Last edited:
This one with the framing better matched and more shadow areas to compare. All three images below are ETTR and the shadows raised in post-processing.

[IMG width="400px" alt="100% crops from 8MP downsample. Click "original size" to view the crops at full-size."]http://horshack.smugmug.com/photos/i-HbJf8Mv/0/O/i-HbJf8Mv.jpg[/IMG]
100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)
...but all this is just a dream. The single D750 image is considerably better than the 20 exposure stack. There is no way to get around the sensor size advantage regarding DR. It's not working in reality, not even in your examples. This is clearly visible even in the tiny post stamp size image above, and even more clearly at "Original" size.
Stacking reduces image noise by the sqrt(# iamges), so a 20-image stack reduces noise by 3.5EV across the entire tonal range. DxO measures the D750 as having a 2.8EV SNR advtange (midtone noise) and 3.8EV DR advantage over the J4. After stacking 20 images the J4 has a 2/3EV SNR advantage over the D750 and is within 0.3EV of the D750 for DR, since engineering DR is increased by 3.8EV on the shadow end; photographic DR is increased even more on the J4 relative to the D750. Both the math and the images support the conclusion that the J4 matches FF noise performance when stacking images.
Yes well, that is all theory, assuming is true, is only applicable on the same sensor. Besides, your 20 stacked image lost some detail which I assume happened because of the slight movement while you took those. Anyway, I don't understand how you can say that your images support the theory. The single D750 image is definitely better than the stacks.
 
I recently acquired a J4 and a preliminary workflow for producing full-frame equivalent IQ (in terms of noise). There are two main aspects:

ETTR (Expose to the Right)

Exposing the scene so that the brightest highlights wanted/needed are near the rightmost edge of the raw data/histogram. Unfortunately the J4 doesn't provide an RGB histogram and its luminance histogram is imprecise because it's based on a JPEG tone curve (even when shooting raw). I haven't yet tried UniWB or altering the picture control to produce a more reliable histogram. For now I just calibrated the raw levels vs the J4's middle-gray exposure for my most-often shooting situation of landscapes. I've found that middle-gray + 3EV will yield a relatively-safe ETTR exposure, provided the sky is the brightest element of the scene.

When shooting I have the camera's metering set to spot meter and the shooting mode to Creative-Manual. I meter the sky to bring the exposure scale to "0", then adjust the exposure to 3EV brighter than that. This is a little clumsy for repeated shooting but if you're taking multiple shots in the same lighting situation you only need to perform the metering once. A more convenient alternative would be to use Creative-Aperture Priority and keep the exposure compensation pegged to +3EV - that way the camera would always meter +3EV when you point to the sky. Unfortunately the J4 has a bug where exposure compensation to near saturation levels causes the meter to over-exposure by up to 2EV. Not sure if other 1-series bodies have this issue.

Image Stacking for noise reduction

In addition to ETTR I also shoot 20-shot burst for each photograph (you can use a 40-shot on the V3 for even better noise reduction). For this I set the mode dial to "Best Moment Capture" and the capture mode to "Active Selection". For the "Active Selection" option I use the "On release, record preceding frames". That way I can hold the shutter half-pressed and have all 20 frames captured when I release the shutter, producing less motion jerking of the camera. I have the "Capture 20 frames over" parameter set to 1/3 second instead of 1 second - you want to capture all 20 images as quickly as possible because even when using a fast shutter speed the camera is only able to read the sensor at 1/60. I set the "Frames saved(default)" to ALL so the camera defaults to store all 20 images of the burst.

Post-Processing

All 20 images are dropped into Photoshop ACR. I set the WB, sharpening and general adjustments on the first image, then synchronize the other 19 images to the first. I then click "done" and use File -> Scripts -> Load Files Into Stack. I click both "Attempt to Automatically Align Source Images" (since I shoot hand-held there are a few pixels of shift between images sometimes) and "Create Smart Object after Loading Layers". The script takes a few minutes to run. When done, I then go to Layer -> Smart Objects -> Stack Mode -> Median. That takes about 30 seconds. When done you can then crop the image to crop off any blank areas from any pixel shift between the 20 images - I intentionally do this after the Stack Mode because that fills in some of the missing pixels first. Note that a 20-image stack with median averaging reduces noise by 3.5 stops. A 40-image stack (V3) would reduce noise by 5.3 stops. AE bracketing/blending can achieve the same level of noise reduction and only require a few frames but the J-series doesn't support bracketing - plus the stacking technique is more suitable for High ISO hand-held situations as well.

Results

Here are the results of a sample scene. I used a J4 with the 10-30mm PD kit lens and for comparison, a D750 with a Tamron 24-70 VC. I set the aperture, shutter speed, and ISO the same for both bodies. Note that the ETTR on the D750 caused some highlight loss but I wanted to keep the exposure the same so that the noise comparison is fair to the D750. All had the ACR shadow slider increased to max +100 (for the bushes in the shade, simulating a High-DR scene), and all were sharpened in ACR with 35/35 and two passes of USM in PS (30/30/0 and then 300/0.5/0)

[IMG width="400px" alt="100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack."]http://horshack.smugmug.com/photos/i-VFkQHpH/0/O/i-VFkQHpH.jpg[/IMG]
100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack.

And here are full-sized images of the image, downsampled to 8MP:

J4 Matrix-Metered (single exposure)
Is it normal for the J4 to be so noisy at ISO160 and in such good light and ideal conditions? Anyway, this image is overexposed.
Those two D750 images are Spot metered, NOT Matrix. Both are severely overexposed.
The exposure selected on the D750 was based on a matrix-metered reading which was identical between both cameras; the D750 happened to be in spot meter mode when I took the photograph and thus that is what's recorded in EXIF. The J4 is not clipped in the raw data so it is not overexposed; the D750 is clipped per my description to match the exposure of the J4. That was done so that the J4 wasn't given an unfair noise advantage. In the second sample I posted today I dialed back the ETTR on the cameras (to 2.3EV) to keep both within raw clipping.
I am sorry.... you can't do it this way. Also, J4 with a noise advantage over the D750? Are you serious? :-D
 
I recently acquired a J4 and a preliminary workflow for producing full-frame equivalent IQ (in terms of noise). There are two main aspects:

ETTR (Expose to the Right)

Exposing the scene so that the brightest highlights wanted/needed are near the rightmost edge of the raw data/histogram. Unfortunately the J4 doesn't provide an RGB histogram and its luminance histogram is imprecise because it's based on a JPEG tone curve (even when shooting raw). I haven't yet tried UniWB or altering the picture control to produce a more reliable histogram. For now I just calibrated the raw levels vs the J4's middle-gray exposure for my most-often shooting situation of landscapes. I've found that middle-gray + 3EV will yield a relatively-safe ETTR exposure, provided the sky is the brightest element of the scene.

When shooting I have the camera's metering set to spot meter and the shooting mode to Creative-Manual. I meter the sky to bring the exposure scale to "0", then adjust the exposure to 3EV brighter than that. This is a little clumsy for repeated shooting but if you're taking multiple shots in the same lighting situation you only need to perform the metering once. A more convenient alternative would be to use Creative-Aperture Priority and keep the exposure compensation pegged to +3EV - that way the camera would always meter +3EV when you point to the sky. Unfortunately the J4 has a bug where exposure compensation to near saturation levels causes the meter to over-exposure by up to 2EV. Not sure if other 1-series bodies have this issue.

Image Stacking for noise reduction

In addition to ETTR I also shoot 20-shot burst for each photograph (you can use a 40-shot on the V3 for even better noise reduction). For this I set the mode dial to "Best Moment Capture" and the capture mode to "Active Selection". For the "Active Selection" option I use the "On release, record preceding frames". That way I can hold the shutter half-pressed and have all 20 frames captured when I release the shutter, producing less motion jerking of the camera. I have the "Capture 20 frames over" parameter set to 1/3 second instead of 1 second - you want to capture all 20 images as quickly as possible because even when using a fast shutter speed the camera is only able to read the sensor at 1/60. I set the "Frames saved(default)" to ALL so the camera defaults to store all 20 images of the burst.

Post-Processing

All 20 images are dropped into Photoshop ACR. I set the WB, sharpening and general adjustments on the first image, then synchronize the other 19 images to the first. I then click "done" and use File -> Scripts -> Load Files Into Stack. I click both "Attempt to Automatically Align Source Images" (since I shoot hand-held there are a few pixels of shift between images sometimes) and "Create Smart Object after Loading Layers". The script takes a few minutes to run. When done, I then go to Layer -> Smart Objects -> Stack Mode -> Median. That takes about 30 seconds. When done you can then crop the image to crop off any blank areas from any pixel shift between the 20 images - I intentionally do this after the Stack Mode because that fills in some of the missing pixels first. Note that a 20-image stack with median averaging reduces noise by 3.5 stops. A 40-image stack (V3) would reduce noise by 5.3 stops. AE bracketing/blending can achieve the same level of noise reduction and only require a few frames but the J-series doesn't support bracketing - plus the stacking technique is more suitable for High ISO hand-held situations as well.

Results

Here are the results of a sample scene. I used a J4 with the 10-30mm PD kit lens and for comparison, a D750 with a Tamron 24-70 VC. I set the aperture, shutter speed, and ISO the same for both bodies. Note that the ETTR on the D750 caused some highlight loss but I wanted to keep the exposure the same so that the noise comparison is fair to the D750. All had the ACR shadow slider increased to max +100 (for the bushes in the shade, simulating a High-DR scene), and all were sharpened in ACR with 35/35 and two passes of USM in PS (30/30/0 and then 300/0.5/0)

[IMG width="400px" alt="100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack."]http://horshack.smugmug.com/photos/i-VFkQHpH/0/O/i-VFkQHpH.jpg[/IMG]
100% crops from 8MP downsample. The "Matrix" images were for the normal exposure that the camera would choose for matrix metering - in this case 1/1000. The spot-metered ETTR exposures were 1/160. Top row contains shadows (raised +100 in ACR), middle row has midtones and upper midtones, and bottom row is the sky. Click "Original Size" to see the full-size of these crops. "J4 Matrix", "J4 ETTR", "D750 Matrix", and "D750 ETTR" show the results from a single raw; the other two J4 images show the results for a 20-image stack.

And here are full-sized images of the image, downsampled to 8MP:

J4 Matrix-Metered (single exposure)
Is it normal for the J4 to be so noisy at ISO160 and in such good light and ideal conditions? Anyway, this image is overexposed.
Those two D750 images are Spot metered, NOT Matrix. Both are severely overexposed.
The exposure selected on the D750 was based on a matrix-metered reading which was identical between both cameras; the D750 happened to be in spot meter mode when I took the photograph and thus that is what's recorded in EXIF. The J4 is not clipped in the raw data so it is not overexposed; the D750 is clipped per my description to match the exposure of the J4. That was done so that the J4 wasn't given an unfair noise advantage. In the second sample I posted today I dialed back the ETTR on the cameras (to 2.3EV) to keep both within raw clipping.
I am sorry.... you can't do it this way. Also, J4 with a noise advantage over the D750? Are you serious? :-D
Can't do it what way? I'm happy to debate the technique and its results but you're going to need to present objective technical arguments rather than just incredulity.
 
This one with the framing better matched and more shadow areas to compare. All three images below are ETTR and the shadows raised in post-processing.

[IMG width="400px" alt="100% crops from 8MP downsample. Click "original size" to view the crops at full-size."]http://horshack.smugmug.com/photos/i-HbJf8Mv/0/O/i-HbJf8Mv.jpg[/IMG]
100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)
...but all this is just a dream. The single D750 image is considerably better than the 20 exposure stack. There is no way to get around the sensor size advantage regarding DR. It's not working in reality, not even in your examples. This is clearly visible even in the tiny post stamp size image above, and even more clearly at "Original" size.
Stacking reduces image noise by the sqrt(# iamges), so a 20-image stack reduces noise by 3.5EV across the entire tonal range. DxO measures the D750 as having a 2.8EV SNR advtange (midtone noise) and 3.8EV DR advantage over the J4. After stacking 20 images the J4 has a 2/3EV SNR advantage over the D750 and is within 0.3EV of the D750 for DR, since engineering DR is increased by 3.8EV on the shadow end; photographic DR is increased even more on the J4 relative to the D750. Both the math and the images support the conclusion that the J4 matches FF noise performance when stacking images.
Yes well, that is all theory, assuming is true, is only applicable on the same sensor. Besides, your 20 stacked image lost some detail which I assume happened because of the slight movement while you took those. Anyway, I don't understand how you can say that your images support the theory. The single D750 image is definitely better than the stacks.
The claim in the OP is that this technique will match FF in terms of noise. What noise aspects do you see in the J4 image that underperform the D750 image? Regarding the 20-image stacked detail - what you're seeing is less noise, which can be mistaken for detail. And even though I'm not presenting these as a comparison for acuity/detail the J4 holds itself quite well, esp considering this is a $100 kit lens and a 18MP 1" sensor vs a $1000 lens and 24MP FF sensor. The J4 acuity can be further improved using Super Resolution workflows if one desires, such as with programs like Photo Acute.
 
Last edited:
Agree the noise level is a lot lower with stacking but then there is a lot of lost in detail. I have to say having an image free of noise is all very well but lost in detail is quite another. Personally I'd much prefer to have more detail with a bit of noise rather then noise free with a lot of detail lost. The stacking method is interesting though, seen quite a few post using this method to reduce noise and get better DR, but as far as noise is concerned, I'd stick with post noise reduction + unsharp mask for better balance between noise and detail.
 
Agree the noise level is a lot lower with stacking but then there is a lot of lost in detail. I have to say having an image free of noise is all very well but lost in detail is quite another. Personally I'd much prefer to have more detail with a bit of noise rather then noise free with a lot of detail lost. The stacking method is interesting though, seen quite a few post using this method to reduce noise and get better DR, but as far as noise is concerned, I'd stick with post noise reduction + unsharp mask for better balance between noise and detail.
Hmm, I'm not seeing the loss of detail. I see a loss of noise that gives the appearance of loss of detail. If I add 2% gaussian noise back into the 20-image photo then it looks like what might be perceived as extra detail vs the single-J4 image.
 
Last edited:
There's no built-in profile for the Nikon 1 cameras so I used the D800E and Sigma 24-105mm profile, with LR/ACR-exported TIFFs as the input into PhotoAcute.

i-2xhJW5m.jpg


100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)

J4 8MP (20-image stack in PhotoAcute)
 
Last edited:
Agree the noise level is a lot lower with stacking but then there is a lot of lost in detail. I have to say having an image free of noise is all very well but lost in detail is quite another. Personally I'd much prefer to have more detail with a bit of noise rather then noise free with a lot of detail lost. The stacking method is interesting though, seen quite a few post using this method to reduce noise and get better DR, but as far as noise is concerned, I'd stick with post noise reduction + unsharp mask for better balance between noise and detail.
Hmm, I'm not seeing the loss of detail. I see a loss of noise that gives the appearance of loss of detail. If I add 2% gaussian noise back into the 20-image photo then it looks like what might be perceived as extra detail vs the single-J4 image.
The lost of detail I was refering to was more to do with the over exposure then stacking. This is especially apparent in the roof top picture. In that particular picture, the matrix metering looked a lot better then the ETTR method to me. There is less over exposure, and less colour lost. Maybe the noise added to the texture of the wall and tiles as well, but from exposure point of view, it is definitely more accurate. But of cause, the noise level is a lot higher. Not saying ETTR is a poor method, just don't thinkn it is the do all, end all method. I do think under certain condition, this method would work well. As to the stacking, unless you're capturing a 100% static subject, you're always in danger of subject movement, due to for example wind, hand shaking or other factors, which could result in a softer picture.
 
This one with the framing better matched and more shadow areas to compare. All three images below are ETTR and the shadows raised in post-processing.

[IMG width="400px" alt="100% crops from 8MP downsample. Click "original size" to view the crops at full-size."]http://horshack.smugmug.com/photos/i-HbJf8Mv/0/O/i-HbJf8Mv.jpg[/IMG]
100% crops from 8MP downsample. Click "original size" to view the crops at full-size.

J4 8MP (single exposure)

J4 8MP (20-image stack)

D750 8MP (single exposure)
...but all this is just a dream. The single D750 image is considerably better than the 20 exposure stack. There is no way to get around the sensor size advantage regarding DR. It's not working in reality, not even in your examples. This is clearly visible even in the tiny post stamp size image above, and even more clearly at "Original" size.
Stacking reduces image noise by the sqrt(# iamges), so a 20-image stack reduces noise by 3.5EV across the entire tonal range. DxO measures the D750 as having a 2.8EV SNR advtange (midtone noise) and 3.8EV DR advantage over the J4. After stacking 20 images the J4 has a 2/3EV SNR advantage over the D750 and is within 0.3EV of the D750 for DR, since engineering DR is increased by 3.8EV on the shadow end; photographic DR is increased even more on the J4 relative to the D750. Both the math and the images support the conclusion that the J4 matches FF noise performance when stacking images.
Yes well, that is all theory, assuming is true, is only applicable on the same sensor. Besides, your 20 stacked image lost some detail which I assume happened because of the slight movement while you took those. Anyway, I don't understand how you can say that your images support the theory. The single D750 image is definitely better than the stacks.
I'm not seeing what you are seeing. Where is the lost details?

That said I don't like the harsh/sharpened look of the images.
 

Keyboard shortcuts

Back
Top