Navigation

October 28, 2009

Lumolabs: Pentax K-x vs. K-7, sensor and video


Pentax has recently released a new SLR, the Pentax K-x. And while it is positioned at the entry-level market and very competitively priced, it is yet rumoured to have a very good noise performance.

After having done all my lab tests, I may say that the rumors are not true.

The truth is that the K-x may be the 35mm SLR camera which has the best high ISO noise and dynamic range performance to date. This isn't "very good". This is outstanding!

This is my concluding summary which is based on the following individual articles:



Noise and Dynamic Range comparison:

At the top of the article is a compilation of various gray sample patches for both cameras, for direct comparison. The patches are from linear raw files (cf. "further reading" at the end of the article).

The resulting 18% gray level chart is like this:


Signal to noise ratio of 18% gray luminosity and color temperature 2900°K (halogen tungsten). Using manufacturer ISO stops 100, 200, 400, 800, 1600, 3200, 6400, 12800. The interpretation of results is the same as for the corresponding graphs on dxomark.com. E.g., print-normalization is for 8 megapixels. The dotted lines indicate the slope for pure photon shot noise.

Obviously, the K-x has no true ISO 100 and both cameras reduce noise even in RAW at ISO 3200 and higher. The ISO sensitivity of the K-x seems to be slightly higher, like at ISO 100, true ISO were more like 87 for the K-7 (DxO's value for the K20D) and 105 for the K-x. However, the above curves have not been left/right-shifted to take this into account.

Gray level noise at ISO 1600/3200 for the K-x is about the same as at ISO 1000 for the K-7. While gray level noise at ISO 1600 and 3200 are about the same for the K-x, there is a minimal loss of resolution when looking at the pattern part in the full test charts. Nothing serious, though.


Signal to noise ratio (SNR) for the Pentax K-7 and K-x. Full measurement for various luminosities and ISO settings at color temperature 2900°K. Measurement similiar to dxomark.com. The dotted lines indicate the slope for pure photon shot noise. The flattening at 30% gray and brighter is believed to be due to systematic measurement errors like imperfect gray patches. The dashed curves for the K-x have been properly normalized to match the higher sensor resolution of the K-7.

The curve for K-7 only and test images are at:
Full SNR Pentax K-7

The curve for K-x only and test images are at:
Full SNR Pentax K-x

The K-x/ISO 200 and K-7/ISO 100 curves are almost identical down to 1% luminosity.

The full SNR graph clearly exhibits the true strength of the K-x sensor: it maintains the good SNR at low luminosity levels. This is an indication that the read-out noise from this sensor is very low.

Therefore, the K-x sensor outperforms the K-7 in low light situations whereas the K-7 outperforms the K-x in good light. This is further confirmed by the better resolution and more artifact-free images from the K-7.


Video:

The quality from HD video is very good. The resolution is as high as the 720p mode allows. It somewhat more moiré and color moiré artifacts than the K-7 and therefore, may not exactly match its analog and film-like quality in 720p mode. But it is very good nevertheless.


Conclusion:

The Pentax K-x has a sensor which rivals the best. That Pentax makes it available in the entry-level class is a good thing. On the other hand, the Pentax K-7 excells in many respects and even its noise performance isn't bad. It is in line with the rest of the competition which is beaten by the K-x in this respect now.


Pentax K-x: "Editor's choice" (entry level, available light)
Pentax K-7: "Editor's choice" (semi professional)



Further reading: Lumolabs testing methodology.

Lumolabs: Pentax K-x sensor quality


The Pentax K-x is an incredible machine for an entry level SLR. Here, I will look at the image quality which its sensor can produce.

In my lumolab testings, I will not report about ergonomics or lack of features etc. The internet is full of this easily obtained information and I will refrain from duplicating it here.


Test chart results:

The photo at the top of the article is an ISO test shot taken at ISO 1600 and color temperature 2900°K. It is a linearly converted RAW file (cf. "further reading" at the end of the article). Go to the gallery (click the image) to have a look at all test shots, available at original full size.


ISO resolution test chart using FA 31 Ltd. at f/4.5. The inner part is 4x. The Nyquist limit is at "7.1" in the inner part. Watch the original at 100% size.

The K-x resolves down to the Nyquist limit. However, it exhibits color moiré and false demosaicing at the limit frequency. The anti alias filter is very weak or absent as it allows for moiré down to three times the Nyquist limit. Color moiré is visible for textures at the Nyquist frequency.

E.g., note that the "7" patch has false demosaicing and strong color moiré. The same is visible from the zone plate chart. The K-7 doesn't show this in comparison.


Zone plate test chart using FA 31 Ltd. at f/4.5. The big circles have a resolution limit of 1280 LW/PH. The smaller ones are 2x and 4x. The 4x circles at above ~4000 LW/PH have uncolored printing moiré. The Nyquist limit is at 2848 LW/PH which is about 55% into the smallest circle. Watch the original at 100% size.

The zone plate chart of the K-x has false colors (fringing) already at half the Nyquist frequency and "green blobs" at three-quarter of the Nyquist frequency, followed by heavy color moiré at the Nyquist frequency exactly. The latter two effects are limited in spatial frequency space but the fringing is kind of disappointing.

While most testers will not notice the effect and welcome "the good resolving power of the K-x", which "leaves no resolution advantage to the K-7" (anticipated quotes from future pseudo tests), I personally would have preferred a stronger anti alias filter.


Numerical lab result:

The numerical evaluation of all noise charts (the full SNR plot) is as follows:


Signal to noise ratio (SNR) for the Pentax K-x. Full measurement for various luminosities and ISO settings at color temperature 2900°K. Measurement similiar to dxomark.com. The dotted lines indicate the slope for pure photon shot noise.

It is possible to derive both dynamic range and noise from the graph. If extrapolating the 0db point for ISO 100 at luminosity 0.01%, then the resulting dynamic range is 13.3 EV (print-normalized to 13.6 EV). E.g., DxO tests the print-normalized dynamic range of the K20D to be 11.05 EV and of the outstanding Nikon D3X to be 13.65. The flattening at 30% gray and brighter is believed to be due to systematic measurement errors like imperfect gray patches.

The theoretical photon noise separation between neighboring ISO lines is 3dB. There are three effects which can be observed:

- The SNR at ISO 100 is reduced for luminance levels above ~3% and drops to the ISO 200 values. A sign that ISO 100 is no native IO value for this sensor. On the other hand, it is exceptionally high for luminance levels below ~3%, with a boost more like ISO 75 than ISO 100.

- The ISO 3200 line is only 0.5 dB separated from the ISO 1600 line. DxO labs found that this is due to noise reduction applied to RAW data at ISO 3200 and higher. Without such noise reduction, the SNR would obviously be about 2.5 dB lower. This smoothing is even stronger than for the K-7.

Overall, the gray level noise (the SNR at luminosity 18.00%) is very close to the competition (worse at ISO 100, same at ISO 200 and better beyond) while the black level noise (the SNR at luminosity 0.10%) is very small. The ISO 1600 curve provides about the same black signal than ISO 100 pushed +4EV.

I have not measured SNR for luminance below 0.04%. But from extrapolation one may deduce that the K-x has a dynamic range of about 13.6 EV which would be even an excellent value for a full frame camera.


Update - The K-x "bump" at 2%, ISO 100 (2009, October 29):

Because of the importance of the ISO 100 SNR curve at low luminosities for the outstanding dynamic range claim made for the K-x, I have evaluated available testing material again, with special emphasis on this bump.


Signal to noise ratio (SNR) for the Pentax K-x. Full measurement for various luminosities and ISO 100 at color temperature 2900°K. Measurement similiar to dxomark.com. The dotted lines indicate the slope for pure photon shot noise.

This graph is meant to study the "bump" at about luminosity 2% in the ISO 100 curve of the full SNR graph (cf. above). The full 12 EV test chart is compiled from two parts: one exposed normally, another underexposed by 5 stops.

The above plot is a reexamination of both parts, now separately evaluated. There indeed is a discontinuous step (by +1.8 dB) when going from the normal to the -5EV curve which I cannot currently explain. The flattening of the brightest parts looks the same and could indeed be due to imperfect bright print patches. This would imply an underestimation of the "bump" by another 1.7 dB leading to a possible overall overestimation of SNR by +3.5 dB in the dark part.

Even taking this correction into account though, the above plot still confirms an extrapolated 0 dB SNR point at 0.01% luminosity. Therefore, the conclusions made in the other sesctions are not altered by this detail examination.

(end of update)


Conclusion:

- Resolution: Full Nyquist 12.2 Mpixel resolution, some moiré, some color moiré, some demosaicing artifacts. Very weak or absent anti alias filter.

- Noise: Gray level 40.0 dB at ISO 100, 31.6 dB at ISO 1600, dynamic range (print-normalized) >13 EV.

The image quality is good at low ISO values and very good if shadows are pushed-processed. It is outstanding at high ISO values and sets the bar for 2010. I haven't tested the Nikon D700 but from the known results, I say that the Pentax K-x is seriously challenging the D700 in terms of high ISO noise and dynamic range. With a much smaller body, a smaller sensor and a much smaller price tag ... and video ;)

I award the Pentax K-x the following title:

"Dynamic range champion 2009 in the 35 mm body class".



Further reading:
- Lumolabs: Pentax K-7 sensor quality
- Lumolabs testing methodology.

Lumolabs: Pentax K-7 sensor quality


I have previously reported about the image quality from the Pentax K-7.

The corresponding blog articles are here:


Here, I will set a level playing ground for future comparisons of the image quality obtained from various cameras.

In my lumolab testings, I will not report about ergonomics or lack of features etc. The internet is full of this easily obtained information and I will refrain from duplicating it here.


Test chart results:

The photo at the top of the article is an ISO test shot taken at ISO 1600 and color temperature 2900°K. It is a linearly converted RAW file (cf. "further reading" at the end of the article). Go to the gallery (click the image) to have a look at all test shots, available at original full size.


ISO resolution test chart using FA 31 Ltd. at f/4.5. The inner part is 4x. The Nyquist limit is at "7.8" in the inner part. Watch the original at 100% size.

The K-7 resolves down to the Nyquist limit with very little color moiré and false demosaicing. The anti alias filter is rather weak as it allows for moiré down to twice the Nyquist limit. Color moiré is under control though.


Zone plate test chart using FA 31 Ltd. at f/4.5. The big circles have a resolution limit of 1280 LW/PH. The smaller ones are 2x and 4x. The 4x circles at above ~4000 LW/PH have uncolored printing moiré. The Nyquist limit is at 3104 LW/PH which is about 61% into the smallest circle. Watch the original at 100% size.

The transition at the Nyquist limit is rather smooth (the printing moiré being stronger than the sensor moiré) with almost no color moiré.


Numerical lab result:

The numerical evaluation of all noise charts (the full SNR plot) is as follows:

Signal to noise ratio (SNR) for the Pentax K-7. Full measurement for various luminosities and ISO settings at color temperature 2900°K. Measurement similiar to dxomark.com. The dotted lines indicate the slope for pure photon shot noise.

It is possible to derive both dynamic range and noise from the graph. If extrapolating the 0db point for ISO 100 at luminosity 0.07%, then the resulting dynamic range is 10.5 EV (print-normalized to 10.9 EV). E.g., DxO tests the print-normalized dynamic range of the K20D to be 11.05 EV which is about the same. The flattening at 30% gray and brighter is believed to be due to systematic measurement errors like imperfect gray patches.

The theoretical photon noise separation between neighboring ISO lines is 3dB. There are two effects which can be observed:

- The SNR drops faster than required by photon shot noise at gray levels under about 1%. This reduces the dynamic range. Gordon B Good found that this is due to a rather large read-out noise caused by the variable gain amplifiers (VGA) integrated into the CMOS sensor chip.

- The ISO 3200 line is only 1 dB separated from the ISO 1600 line. DxO labs found that this is due to noise redauction applied to RAW data at ISO 3200 and higher. Without such noise reduction, the SNR would obviously be about 2 dB lower.

Overall, the gray level noise (the SNR at luminosity 18.00%) is very close to the competition while the black level noise (the SNR at luminosity 0.10%) is rather large. This is of concern if brightness needs to be pushed in post processing. Interestingly, the ISO 1600 curve provides a better black signal than ISO 100 pushed +4EV.


Conclusion:

- Resolution: Full Nyquist 14.6 Mpixel resolution, weak moiré, very weak color moiré, nearly no demosaicing artifacts.

- Noise: Gray level 40.5 dB at ISO 100, 28.1 dB at ISO 1600, dynamic range (print-normalized) 10.9 EV.

The image quality is excellent at low ISO values and if shadows aren't pushed-processed, good at low ISO values otherwise and in line with the 2008 competition at high ISO values.


Further reading: Lumolabs testing methodology.

Lumolabs: Pentax K-x HD video quality


The Pentax K-x is an incredible machine for an entry level SLR. Here, I will look at the quality one can expect to obtain from its 720p@24Hz HD video capture mode.

In my lumolab testings, I will not report about ergonomics or lack of features etc. The internet is full of this easily obtained information and I will refrain from duplicating it here.

The above is a frame from a dumb sample video, available here:



Test chart results:


Video frame from 720p HD mode. ISO resolution test chart using FA 31 Ltd. at f/4.5. The inner part is 4x. The 720p Nyquist limit is at "8.5" in the outer part.

Watch the original at 100% size.


Zone plate test chart. The big circles have a resolution limit of 1080 LW/PH. The smaller ones are 2x and 4x. The Nyquist limit is at 720 LW/PH which is about 67% into the big circle. Watch the original at 100% size. The sampling frequency is clearly visible by the position of the four false color disks positioned at an 70% radius. The false color at and remaining artefacts beyond the sampling frequency is due to a missing video anti alias filter.

Note that the false color artifacts are significantly more intense than from the K-7, overall and esp. at 45°.

A similiar image for the Canon 5DmkII is visible here: Canon 5DmkII zone plate test.


Rolling Shutter Test:


Panning left and right to evaluate the rolling shutter effect.

Note that the sensor of the Pentax K-x is read out in the opposite direction than that of the Pentax K-7, like bottom-up rather than top-down.

In comparing with the K-7, please note that the panning speed is not exactly equal. The magnitude of the rolling shutter effect is proportional to the panning speed. Also note that the K-x video has slightly more blur. Lightning condition and aperture has been the same, though.

Some testers have stated that the K-x has a stronger rolling shutter effect than other cameras. I cannot confirm this. Maybe, the opposite direction of skew disturbed the viewing habits of video testers ...


Conclusion:

The Pentax K-x delivers very good HD video quality. It has color moiré artifacts close to the Nyquist frequency which are due to subsampling without an anti alias filter. The artifacts are strong and not so nicely confined to a close neighborhood around the sampling frequency. So, in real footage one has to expect to see some textures with color moiré. This is a problem of all SLRs though, specifically those sampling at natively 720p.

The effect seems to be about that of a Pentax K-7 along horizontal and vertical lines. But the K-7 has no such artifacts along diagonal lines which may point to the fact that it uses more and differently aligned pixels in its subsampling procedure. Obviously, the K-7 scans 1024 lines (using 864 in 16:9 format) whereas the K-x only scans 720 lines.

Interestingly and unlike the K-7, the K-x has no magenta/green fringing and no double seam effect along near vertical lines. The subjective effective resolution in 16:9 is about 1350x717 or 1.0 MPixel. This is an excellent value not achieved by consumer HD camcorders. It even surpasses the 1280 limit ;) The K-7 has slightly less resolution in the horizontal direction but much better resolution along diagonal lines.


Further reading:
- Lumolabs: Pentax K-7 HD video quality
- Lumolabs testing methodology.

Lumolabs: Pentax K-7 HD video quality


I have previously reported about the HD video mode in the Pentax K-7.

The corresponding blog articles are here:


Here, I will set a level playing ground for future comparisons of the video image quality obtained from various cameras.

In my lumolab testings, I will not report about ergonomics or lack of features etc. The internet is full of this easily obtained information and I will refrain from duplicating it here.

The above is a frame from a dumb sample video, available here:
Bavarian Camels (K-7 version).


Test chart results:


Video frame from 720p HD mode. ISO resolution test chart using FA 31 Ltd. at f/4.5. The inner part is 4x. The 720p Nyquist limit is at "8.5" in the outer part. The 1024p Nyquist limit is at "10.0" in the outer part. The color moiré shows that the 720p frames are downsampled from 1024p video frames.

Watch the original at 100% size!


Video frame from 1536x1024p HD mode. The 1024p Nyquist limit is at "10.0" in the outer part and "2.5" in the inner part. The resolved parts with distinct lines is about "8.0" in horizontal and "8.5" in vertical direction, corresponding to 1200x850 pixels in 3:2 or about 1200x717p pixels in 16:9 mode.


Video frame from 720p HD mode. The big circles have a resolution limit of 1080 LW/PH. The smaller ones are 2x and 4x. The Nyquist limit is at 720 LW/PH which is about 67% into the big circle. The Nyquist limit of the 1024p sampling is at 864 LW/PH which is 80% into the big circle. Watch the original at 100% size.
The sampling frequency is clearly visible by the position of the four false color disks positioned at an 80% radius. The false color at and remaining artefacts beyond the sampling frequency is due to a missing video anti alias filter.

A similiar image for the Canon 5DmkII is visible here: Canon 5DmkII zone plate test.


Rolling Shutter Test:


Panning left and right to evaluate the rolling shutter effect.

Note that the sensor of the Pentax K-7 is read out in the opposite direction compared to that of the Pentax K-x, like top-down rather than bottom-up.


Conclusion:

The Pentax K-7 delivers stunning HD video quality, specifically in 720p. It has color moiré artifacts close to the Nyquist frequency which are due to subsampling without an anti alias filter. The artifacts are strong but seem to be nicely confined to a close neighborhood around the sampling frequency. So, in real footage one has to really hit the "wrong" texture to see the effect.

The effect seems to be about that of a Canon 5DmkII along horizontal lines. But the Canon has no such artifacts along vertical lines, pointing to supersampling within lines and line skipping. The Pentax K-7 does line and row skipping. Obviously, the 5DmkII scans 1053 lines whereas the K-7 only scans 1024 lines.

Moreover, in 1024p along nearly vertical edges (cf. the 1024p resolution test chart), one can see the magenta/green fringing and a double seam effect. This is much less visible in 720p though. Additionally, it may be possible to cure it in post processing. The subjective effective resolution in 16:9 is about 1200x717 or 0.9 MPixel. This is a good value actually not achieved by consumer HD camcorders.

Due to its supersampling applied to 720p video, footage from the K-7 looks more cinema-like and analog than footage from entry-level dSLRs.


Further reading: Lumolabs testing methodology.

Lumolabs: Welcome and testing methodology


Welcome to Lumolabs.

I never was really satisfied with the big lab tests (dxomark.com and dpreview.com) and really was disappointed by print magazine and other ezine tests. So, I decided to set up my own testing laboratory, based on experience as a photographer, insight as a physicist, software writing and image processing skills and patience.

For the time being, I am not challenging the big labs and confine my ambitions to the following areas:

1. RAW file image quality, in particular noise, dynamic range and resolution.
2. Video frame image quality, in particular resolution and artifacts.
3. Pentax.

I am not looking into color accuracy (not even chrominance noise...) or the camera's JPG engine. Or autofocus, ergonomy, build quality, etc. I may do lens tests and other vendors later as well.


Let me provide you with a short summary of my testing methodology:


RAW file treatment

I use the RAW file from the camera (DNG preferred) and use Adobe Lightroom 2.4 (LR) to do the demosaicing. The standard settings must be altered in the follwing ways to get a neutral demosaiced 16Bit sRGB TIFF file:

- Gray card calibrated white balance (I use 2900°K halogen tungsten light).
- Blacks 0 (changed from 5)
- Brightness 0 (changed from 50)
- Contrast 0 (changed from 25)
- Sharpening 0 (changed from 25)
- Noise reduction Color 0 (changed from 25)

I then read the resulting 16Bit sRGB TIFF file into my own lumolabs software. I properly convert the sRGB response curve (which isn't exactly a gamma curve) into linear colors and 18% gray is RGB 117.6/255.


Taking test chart photos

I then take test chart photos with controlled manual settings, including manual focus. About the slowest shutter speed is 1/15s.

Some think that noise test shots must be taken in low tungsten light. I agree for the tungsten part and I use it. I disagree for the low light part, though. A short exposure is a perfect approximation to low light because the read-out time of the sensor is longer than 0.1 seconds anyway whatever fast be the shutter. So, only if I wanted to study exposure time longer than, say 1/5s would low light be able to have an influence. In consequence, I don't publish exposure times shorter than 1/10s. Additionally, except for initial focus, I avoid using live view in noise tests to not overheat the sensor.

One of my test charts is the zone plate chart useful to evaluate anti alias filter and Bayer mosaic filter artifacts. It is shown in the beginning of the article. It is allowed to download it for own testing purposes, providing proper credit is given.

Other testing charts are an ISO-12233 resolution chart which I modified to include inner parts of 4x the resolution. Another chart is the noise testing chart with defined gray and color patches. All charts are printed on A2 paper using a high end photo printer. There are minimal printing artifacts visible at and beyond 4000 LW/PH (line widths per picture height). They do not disturb the testing results though.

The density range of the printed noise test chart is about 7 EV. So, I take a second shot at -5EV and obtain a combined test chart of about 12 EV dynamic range (luminance range 0.02% to 100%). Which immediateley turned out to be outperformed by the recent Pentax K-x ;)

My software will search for patches of minimal variance to exclude the effect of scratches and the like which may be present on the printed test chart. I am not sure though that there aren't some invisible defects below the -40dB range. They do not disturb the testing results though, except maybe for very bright patches.


SNR, noise and dynamic range

I am unrelated to DxO Labs, Boulogne, France. However, whereever I found their methodology to be appropriate (cf. dxomark.com), I adopted it. I disagree on some of their methodologies but some are best practise to be followed. My biggest concern with DxO is their lack of examination of influence of spatial frequency dependencies. They only "correct" their results for resolution effects as to be expected from pure photon shot noise. IMHO, this isn't good enough. I am currently working at a new testing methodology to overcome this. For the time being though, I am happy to adopt DxO's pixel-level measure methodologies.

It is therefore important to understand the SNR [dB] figure. It is explained here: SNR.

Another important detail is the print normalization: Noise, dynamic range etc. are all improved by downsizing an image. Still, measures take place at the varying resolution of the sensor. Like DxO, I use a "print-normalization" calibrated to a 8 MPixel resolution. The normalization formulae are here: print-normalization.

SNR [dB] can be easily understood by using the follwing hints:
- A patch of basically random pixels has an SNR of 0 dB.
- A decrease of light by 1 EV, doubling the ISO, going to a 50% darger region in an image or doubling the number of pixels all amount to a change of SNR by -3dB.
- Going from 1.53 crop APS-C to full frame amounts to a change of SNR by +3.7dB. So, 3dB is a lot ...

My full SNR curves can be directly compared with the DxO mark tab "Full SNR". Note however that DxO uses noise patches made from glass, has a different light source most importantly at a different color temperature and uses a different raw converter. Also, in derived measures, I don't include ISO sensitivity variance and possible raw data noise reduction (smoothing).

The dynamic range is the luminance level where SNR reaches 0dB, converted to f-stops and normalized for resolution. Note that the 0dB level is arbitrarily choosen by DxO and is resolution dependent. They formula-correct for resolution dependency but fail to understand that the full dynamic range must be measured at a much lower resolution. I plan to address this issue.


Video frame tests

The video frames studied in the video section are extracted using the Quicktime Pro "export as an image" feature. Because the video uses the camera's JPG engine, here are the parameters:

- Pentax: WB manual (2900°K), "Natural color profile preset", all other parameters at default values (like noise reduction and sharpness -1).


Enjoy the tests :)