I’ve tested this preset with some Canon cameras and it was quite close (although in theory it should be the same), but there is more accuracy with Nikon D3/D300 cameras. If there is going to be demand for this preset I’ll try to produce a more accurate Canon version. Please do let me know how this preset works for you.

m8-on-nikonlrtemplate.zip
(more…)

Dynamic range

A desired improvements is more dynamic range. To test this we photographed a special 41 transparent step chart with 13.6 EV stops of dynamic range. We define dynamic range as the number of steps between maximum saturation where the sensor can no longer capture light and the floor noise at the 1db limit. This means that dynamic range is limited by noise. We estimate a margin error of 0.15EV in our measurements.

The chart bellow shows the dynamic range in EV of each channel in each of the ISO settings.

As you can see, the green channel, which is the main channel, has about 0.3EV more dynamic range at 200ISO, the base ISO setting for the Nikon D300. It is interesting to note that the Nikon D200 gains some dynamic range at 200ISO in the green channel but loses some in the red channel. Comparing the base ISO setting of both cameras, the Nikon D300 gains 0.6EV more dynamic range in the green channel, which is not bad at all.

Because noise limits dynamic range it is no surprise that the Nikon D200 has only 7.6EV of dynamic range at 1600ISO while the Nikon D300 shows an excellent result of 8.6EV.

As for the long debate of 14bit vs. 12bit, we can clearly see a small advantage to having more bits in dynamic range. Sometimes having more accuracy means better signal to noise ratio at the low signal parts of the image.

Signal to noise ratio

To test signal to noise ratio we used the same data from the previous test. This time however, we calculated the signal to noise ratio from the 2 brightest stops of the chart. The results represent the ability of the sensor to gather light and are less relevant to the electronic side effect noise of the image. This is because the sensor is overwhelmed with photons and the extra electronic charge has very little effect on the overall charge.

The higher the result the better the sensor is capable of capturing light. The pixel size of the Nikon D300 is approximately 5.54 µm while the pixel size of the Nikon D200 is 6.09 µm. That alone should give the Nikon D200 slightly better performance. But there are many more factors to consider, such as the density of the microlenses and the spectral distribution of the color filters on top of each pixel.

The results are very interesting indeed. At seems that in most cases at low ISO settings the Nikon D300 has a better signal to noise ratio, even though the pixel size is slightly smaller. This means that Nikon (or shall I say Sony?) has done a very good job.

Noise floor

Noise floor is calculated by using the same data used in the previous tests. But this time we calculate the standard deviation values at the lower two EV steps. These results indicate the overall performance of the system. In most cases the last two stops contain some signals from light, but also contain some electronic noise that should not have been added to the image. Lower results are better.

From these results there is no doubt that the Nikon D300 is a big improvement on the Nikon D200. This is probably due to Nikon’s better handling of the electronics in the D300 but it is more likely due to the fact that CMOS sensors can separate non photon noise from light signal and subtract that noise from the each pixel. This is something CCD sensors can’t do. The result is a much less noisy RAW file. Once again, we see a small advantage to 14 bit accuracy which reduces the rounding (upwards or downwards) of signal and noise.

Noise reduction improvements

The following is for the JPEG shooters among you. While the improvements above were all about RAW files and sensor SNR, here we will discuss improvements Nikon has made to their noise reduction algorithm. The low SNR and low noise of the Nikon D300gives us a better starting point for noise reduction algorithms, so we will try to highlight some improvements in the Nikon D300 that bothered us in the Nikon D200.

All the crop images below are taken from our studio setup with the same 50mm Nikon lens (at f8) at 1600ISO. All crops are 100% of our studio setup.

Chroma noise and pleaseing grain

One thing the Nikon D200 was free of is chroma noise in flat areas. This is not different in the New Nikon D300. Only when setting the noise reduction setting to the OFF setting there is some chroma noise.

• RGB View
• Luma View
• Chroma View

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Grain is one of the most important aspects of noise reduction - get grain right and people will accept your image as pleasing (something that adobe seems to miss). By clicking the Luma view you can see that the Nikon D300 has a little less visible grain. I do not particular like that, and I much prefer the Low result of the Nikon D300 for grain.

Shadow grain

Grain in the shadows was one of the most problematic issues with the Nikon D200. While most of the tone range filtered well, the dark tones contained too much grain. This made images with mostly dark tones look very grainy and noisy.

• RGB View
• Luma View
• Chroma View

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Here the shadows are filtered much better. This helps the image look more pleasing and appealing overall. I do feel that the Normal NR setting filters a bit too strongly, I much prefer the Low setting as it leaves some nice looking grain which makes the image look a bit more natural.

Details

One of the common misconceptions photographers have about noise reduction is the loss of detail. Performing noise reduction does not necessailry mean losing details. In many cases, details that are otherwise hidden by noise can be “extracted” by detecting fine edges. On the other hand, if the algorithm is incapable of detecting those fine edges some unneeded filtering may occur. In reality, there is always a tradeoff that needs to be made between details and filtering. If the algorithm is very good, this tradeoff is small as it is possible to filter noise and detect fine edges as well. But even then, there will be a small tradeoff which in many cases the camera maker lets the user choose.

• RGB View
• Luma View
• Chroma View

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

The Nikon D300 has a small advantage over the Nikon D200 when it is set to Normal NR. But once again, using the Low setting shows those nice little details. I am quite pleased with the Low setting.

Shadow detail

Details in the shadows is yet another issue that bothers us with the Nikon D200. We are happy to see that Nikon has improved this issue.

• RGB View
• Luma View
• Chroma View

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

As you can see, there is good progress here. The Nikon D300 shows better results in every setting. I prefer the Low setting as it gives good balance between edges and grain.

False colors

False colors are a side affect of noise reduction. Those are little color noise leftovers nikon left behind on edges. I am happy to say there is a good progress on this issue too.

• RGB View
• Luma View
• Chroma View

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

Nikon D200 Normal NR	Nikon D300 Normal NR	Nikon D300 Low NR
Nikon D300 High NR	Nikon D300 OFF NR

As you can see only when the Nikon D300 is set to OFF some false color artifacts appear. There is also very faint false color with the Low setting. This problem seems to be handled very well now.

Final words

Both in our lab tests and in real world the Nikon D300 shows over all less noise than the Nikon D200. This is very true for RAW files that are now more usable then ever. When using JPEG files from of the Nikon D300, we see some very good progress on Nikon’s ability to reduce noise (performance which many different RAW converters can only dream of). The big breakthrough was accomplished with the Nikon D200, but now all of those image processing problems in the Nikon D200 have been solved or improved. The main noise reduction improvement stems from the fact that Nikon can now filter dark tone more strongly than highlights or mid tones.

I much prefer the Nikon D300 1600ISO JPEG results when it is set to Low NR setting. There are more details with not too much extra color noise, but more important the monochromatic grain with the Low setting looks more pleasing. As a results, the whole image looks more pleasing and detailed.

With the Nikon D3/D300 there is a new type of autofocus technology. Nikon has added to the mix a new ingredient: Color.
For quite some time now, Nikon has been using their 1005 pixel sensor system for metering. It works quite well by comparing a database of scenes to the scenes that the photographer is metering.

Now however, Nikon is using the 1005 pixel sensor in combination with the new 51 point autofocus system. Set the camera to 51 points 3D mode, lock one of the points on the object and you are all set. The camera will now remember the scene when the focus point was locked and will try to track the object across the frame.

I have made a short clip so you could see how this works through the viewfinder. You’ll notice the 1005 pixel sensor view at the right. This is a simulation of how the 1005 pixels sensor of the Nikon D300 viewed the scene.

Scene recognition skin tone (face) detection

As an added bonus, Nikon provides a new mode called “Auto Area AF”. When the camera is set to this mode you can’t choose any AF point. The camera automatically chooses the AF points based on a database of scenes Nikon as “taught” the system, using same principle as the metering system. It does not just try to find the closest object.
There is another twist. Because the camera can see color, it is possible to detect skin tone. I’ve made another clip this time to show how the camera automatically selects skin tone when available. It seems that Nikon tries to detect the actual face and eyes as sometimes the focus points are actually on the eyes, depending on the size of the face in the frame.

Final words

We are definitely talking about the future there. There is so much potential in the idea to combine two dimension color sensors and autofocus systems. Just think what could be done with more pixel resolution (face recognition, player number tracking, live view, etc…). I reckon this is where autofocus systems are heading in the future. I can’t wait.

Many of us use RAW these days, and some of us prefer to use DPP due to the excellent picture style system. You should be aware however, that the results you get out of DPP with high ISO images are not as good as the JPEG out of the camera. There is a quick fix you can do to produce better results, but it is really something Canon should address back in Tokyo.

Noise reduction and user controls

DPP allows you to define three intensity levels for both Luma filtering and Chroma filtering separately. By default, these controls are set to OFF, this does not mean there is no noise reduction at all, just that the intensity of filtering is somewhat reduced.

You can set noise reduction to Low and High settings, while there are some advantages to the Low setting in low ISO image, at high and medium ISO images it is best to set both controls to the high setting.

The example below is a 100% crop from our studio scene which we captured using the EOS 1D Mark III at 3200ISO. Move your mouse between the buttons to view the results.

100%
DPP OFF	DPP Low	DPP High

Next to the camera

Most photographers would think that RAW will always produce better image quality than the JPEG file of the camera. In some cases, depending on the RAW software being used, the results are better. The example below shows the difference between the original JPEG out of the EOS 1D Mark III and DPP set to high.

100%
DPP High	EOS 1D Mark III JPEG NR ON

As you can see, DPP is able to produce sharper results than the camera. However, on flat areas (areas that do not contain edges) and dark areas, DPP does not produce results that are as pleasant as the JPEG out of the camera. The flat areas of the image contain many small bright “dots” that disturb the viewer. Those “dots” may produce the illusion that there are more details in the image, especially when viewing the image with small enlargement percentage.

This is only an illusion, as you can see from the example below, DPP does not show more details than the JPEG out of the cameras.
You must distinguish between details and sharpness. A sharper image does not mean it holds more details, it means the large dominant edges are simply sharper.

100%
DPP High	EOS 1D Mark III JPEG NR ON

When looking closely (300%) at the DPP results, you can see that some pixels are brighter than the environment they are in. In other words, they are brighter than the average and some are darker than the average. This phenomenon will be very familiar for those of you who tuned or designed noise reduction algorithms and it is called “filtering holes”. These “holes” are small pixel groups that were not filtered.

300%
DPP High	EOS 1D Mark III JPEG NR ON

This phenomenon takes place when thresholds that are set to distinguish between pixels that need to be filtered and pixels that do not need to be filtered (or filtered differently) are too low.
In this case, if the average of a group of pixels exceeds the threshold value, no filtering will occur. The results below in 300% view shows how well the EOS 1D Mark III is tuned compared to DPP.

Both the noise reduction system and the edge enhancement (sharpening) system could produce such poor results. To understand which system is responsible, we conducted a simple test.

By closing the sharpening setting completely in DPP but leaving the noise reduction settings on high, we took the edge enhancement algorithm out of the equation. We then applied an unsharp mask filter using Photoshop with the following settings:

Amount: 150
Radius: 0.7
Threshold: 0

We applied the same filter on the JPEG out of the EOS 1D Mark III to understand which file was filtered better. You can view the results below.

300%
DPP High	EOS 1D Mark III JPEG NR ON

From these results we conclude that DPPs noise reduction thresholds are probably configured lower than the noise reduction thresholds of the EOS 1D Mark III. Assuming of course that the noise reduction algorithm is the same. Ultimately it is likely that both DPPs edge enhancement and noise reduction thresholds require improvements.

It is to be expected that Canon should try to extract more details or to produce a sharper image with DPP. But I feel that the price Canon and all Canon’s users are paying is too high. I generally prefer the results from the EOS 1D Mark III JPEG file rather than the results from DPP with high ISO images.

Since there is no way for the user to tune thresholds, the best thing we can do for now is to lower the sharpening level to 1 or even 0, while keeping the noise reduction setting on high for high ISO images. By doing so, the results won’t be as sharp as before, but the flat DC areas will appear more pleasing with less filtering holes.

By toggling between the images below you can see that as the sharpening level is reduced, less filtering holes appear.

100%
DPP NR High sharpening 3	DPP NR High sharpening 1	DPP NR High sharpening 0	EOS 1D Mark III JPEG NR ON

Final words and recommendation

It is expected for RAW software to produce more details and sharper edges. But the way this is implemented in DPP is not good enough. There are improvements to be made, and I think Canon can improve the results easily by simply improving the tuning of the algorithms.

Furthermore, I think canon underestimates the importance of defaults values. With the current DPP defaults values, high ISO images are not as good as the EOS 1D Mark III JPEG file. I would like to see DPP configured to its best for each ISO setting and for each camera model. The same way it is done with DIGIC III for each camera Canon has made. The user should then tune the image according to his preferences by using the controls.

Compared with the camera DPP does produce more detail in some very specific areas. It is most likely that the internal sharpening of the EOS 1D Mark III is reduced when the camera is set to high ISO levels. Therefore, by sharpening the JPEG of the EOS 1D Mark III, we can see how well the internal noise reduction is configured, as the images below show.

100%
DPP NR High sharpening 3	EOS 1D Mark III JPEG NR ON + Photoshop USM	EOS 1D Mark III JPEG NR ON

As you can see by toggling between the images, the JPEG file could be sharpened to produce sharper edges but without introducing artifacts at flat areas. This means that the configuration of the noise reduction algorithms inside the EOS 1D Mark III is probably better for flat areas. It’s worth noting that DPP seems to filter Chroma edges slightly better.

Recommendations for Canon

* Configure thresholds and algorithms to exceed or match the image quality of the camera, while producing pleasing grain.
* Configure the default values of both sharpening and noise reduction to match the camera. Do not keep noise reduction levels OFF while the camera clearly is configured differently.
* Improve edge detection in order to extract more details and improve edge sharpening while making sure the flat DC areas are not affected so as to give an unpleasing result.

Recommendation for DPP users

For now, I would recommend to set the sharpening level in DPP to 0 or 1 in order to avoid emphasizing the flittering holes with high ISO images. It is also possible to set noise reduction to HIGH (both) and set the sharpening level to 0. Then, use solutions such as Noise Ninja to “fix” the filtering holes Canon left behind. The last stage should include sharpening using your favorite tool.

See this page to download Focus Finder and for more details.

Download Focus Finder droplet (Mac / PC)

Focus Finder is a Photoshop droplet that can find the focus point in your images. Just drag and drop any image on the droplet and Focus Finder will attempt to find the strongest most detailed part of the image. A red circle will mark the focus point.

Focus Finder was specifically designed to test focus tracking with images taken by the Canon EOS 1D Mark III camera. However, Focus Finder can find the focus point on any 8 megapixel or larger image.

There are a few drawbacks:

With the limited Photoshop tools available it is not possible to determine how sharp the image is. Focus Finder will always find the strongest edges in the image, even if those edges are very weak.

Focus Finder will mark the most detailed part of the image. In some cases where there is a detailed background that is a part of the focus plain, Focus Finder might mark the background as the focus point. This can happen even if the subject is perfectly sharp, but has less detail than the background.

The image above is a good example. The horses head is the subject and it is perfectly sharp. However, the ground is also a part of the focus plane and has much more details than the head. Consequently, Focus Finder has marked the ground as the focus point.

Focus Finder is not very suitable for bird photography, but it might work well when the subject fills the frame.

Guidelines for focus tracking testing with Focus Finder

Make sure your subject has some strong hard edges. For example, if you choose to test your camera with a human subject try to use a shirt that has some large text printed on it. A shirt with many strong textures is preferred.

Do not add artificial pixels (text, squares, marks, etc…) to the image before running Focus Finder.

Try not to include the ground in your images. If the subject has many strong edges Focus Finder might find the subject as well as the ground (As the first example below show).

More examples