K20D burst mode has 768x1024 internal resolution

falconeye · 03-02-2008, 05:24 PM

In thread Fireengine K20D movie in full HD I have posted a fireengine video in full resolution MP4/AVC.

Discussion on a German board have led to the conclusion that the "native" resolution of images captured in burst mode is 768x1024, then scaled up horizontally to 1536x1024 (1.6 MPixel).

Here are two 400% crops from the original images in the video:

Smallest details seem to be 2pixels wide by 1pixel high.

This resolution is still ok for 720p HD video but is not exactly what to expect from the burst mode specification as given by Pentax.

For everybody considering the burst feature, e.g., for sports photography, base your computations on a usable anamorphic resolution of 768 wide by 1024 high. At this resolution, however, the image is sharp at 100% crop and outperforms so-called HD images from consumer HD cameras (which look bad at a 100% crop).

falconeye · 03-02-2008, 05:52 PM

I've done one more test as a logical response to the finding above:

One can treat the images captured in burst mode as follows:

(1) Resample to 768x1024 resolution with NO interpolation
(2) Resample the result to 1536x1024 resolution with bicubic interpolation

The result are much more naturally looking images, the alias artifacts are gone:

I recommend everybody who wants to use the burst feature to treat the resulting images in this way. Sort of Day/Night difference.

However, one artifact isn't removed that way: There is still 2 pixel wide purple fringe on the right side of dark areas of all vertical contrast borders, and green fringe on the left side. This is the same for all locations in the image (left and right of center), so I guess it cannot be the lens. There is no PF issue with horizontal contrast borders, though.

Does anybody know a software which removes this kind of constant PF from an image?

I guess it comes from some strange binning done to get the burst images...

Images from burst mode, when treated both for alias artifacts (cf. above) and the constant PF, are of good enough quality to be useful. But they shouldn't be used w/o further PP.

jsherman999 · 03-02-2008, 05:58 PM

Originally posted by falconeye

I've done one more test as a logical response to the finding above:

One can treat the images captured in burst mode as follows:

(1) Reample to 768x1024 resolution with NO interpolation
(2) Reample the result to 1536x1024 resolution with bicubic interpolation

The result are much more naturally looking images, the alias artifacts are gone.

I recommend everybody who wants to use the burst feature to treat the resulting images in this way. Sort of Day/Night difference.

How does one "Reample the result to 1536x1024 resolution with bicubic interpolation"?

Thanks very much for your analysis, BTW.

.

falconeye · 03-02-2008, 06:15 PM

Originally posted by jsherman999

How does one "Resample the result to 1536x1024 resolution with bicubic interpolation"?

Ok, I assume basic post processing experience here.

In Photoshop CS3, you

use Menu Item "Image -> Image Size ..." (translated back from German, so the exact naming may differ)
type in 768 as new width (change units to pixels and disable proptional sizes first as you aren't changing the height) and use "nearest neighbor" as resampling algorithm.
use the Menu Item a second time, but now
type 1536 as new width and use "bicubic" as resampling algorithm.

I am sure, it is even simpler in Lightroom and can be scripted. And it should always be done when importing bursts as my sample images above may have illustrated.

raz · 03-02-2008, 07:08 PM

I doubt that what you are saying is right, because a 100% increase of resolution from 0.7 Mpixels to 1.5 would be to visible. Can you attach some shots (JPG directly from camera not mp4) of a burst ? I've seen some from others but I canot find the links anymore.

falconeye · 03-02-2008, 07:24 PM

Originally posted by raz

I doubt that what you are saying is right, because a 100% increase of resolution from 0.7 Mpixels to 1.5 would be to visible.

Seeing is believing. Just look at my example above. The simple point is that the 1.6 MPixel images from the Pentax burst mode are already upsampled, but badly so. So, downsampling to recover the internal version and doing it right improves much. Still no 1.6 MPixel sharpness, but now w/o the artifacts. And of course, one must upsample to recover the correct aspect ratio.

Originally posted by raz

Can you attach some shots (JPG directly from camera not mp4) of a burst ?

There is nothing wrong with a frame from a good MP4/AVC. But for added convenience, here is one unprocessed original:[IMGWIDELEFT]http://www.livis.de/download/pub/fireengine.jpg[/IMGWIDELEFT]

raz · 03-02-2008, 09:33 PM

Is this directly from the camera ? What settings did you use for jpgs ? This look very bad, the antialiasing is bad (it may be as you sugest a problem with the resampling...)

I've seen something like this in the jpg of a preproduction model, but on dpreview forum on the ones with firmeware 1.0 I didn't noticed that problem ?

falconeye · 03-03-2008, 05:35 AM

Originally posted by raz

on dpreview forum on the ones with firmeware 1.0 I didn't noticed that problem ?

Unfortunately, you don't provide a source. I could only find one sample on dpreview with 1.00 firmware. It shows a squirrel. It also has this purple/green fringe problem, hence I believe it is taken from a K20D burst. However, the image is soft overall (so no alias artifacts to expect anyway) and moreover, was scaled down 83% at least.

It may be that in this thread here, you see the first serious investigation of K20D burst mode resolution.

Originally posted by raz

Is this directly from the camera ? What settings did you use for jpgs ? This look very bad, the antialiasing is bad (it may be as you sugest a problem with the resampling...)

As I have said before, JPEG directly from the camera (unlike the squirrel on dpreview), my JPEG setting was *** (but I guess it would be ignored anyway). Firmware is 1.00, otherwise you are invited to inspect the EXIF of my sample provided above.

And yes, just trust me. It is exactly as "I sugest". I've invested enough time and have the background to be confident to say so.

But it isn't this bad.

As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?).

What remains just is this thread's topic title:

The K20D's internal burst video resolution is 768 x 1024.

jsherman999 · 03-03-2008, 09:30 AM

Originally posted by falconeye

Unfortunately, you don't provide a source. I could only find one sample on dpreview with 1.00 firmware. It shows a squirrel. It also has this purple/green fringe problem, hence I believe it is taken from a K20D burst. However, the image is soft overall (so no alias artifacts to expect anyway) and moreover, was scaled down 83% at least.

It may be that in this thread here, you see the first serious investigation of K20D burst mode resolution.

As I have said before, JPEG directly from the camera (unlike the squirrel on dpreview), my JPEG setting was *** (but I guess it would be ignored anyway). Firmware is 1.00, otherwise you are invited to inspect the EXIF of my sample provided above.

And yes, just trust me. It is exactly as "I sugest". I've invested enough time and have the background to be confident to say so.

But it isn't this bad.

As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?).

What remains just is this thread's topic title:

The K20D's internal burst video resolution is 768 x 1024.

I concur with falconeye's findings.

.

raz · 03-03-2008, 09:31 AM

Originally posted by falconeye

But it isn't this bad.

As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?).

Yes it is. You get half the resolution, or megapixels from what pentax is saying. Is a huge difference. I thought the 1536x1024 would be a super crisp image taken with full resolution and downsampled or at least taken with 9 super pixels - a group of (3x3). That had a lot of potential, not like a 0.7 megapixel image resized to 1.5.

falconeye · 03-03-2008, 02:46 PM

Originally posted by raz

Yes it is. You get half the resolution, or megapixels from what pentax is saying. Is a huge difference. I thought the 1536x1024 would be a super crisp image taken with full resolution and downsampled or at least taken with 9 super pixels - a group of (3x3). That had a lot of potential, not like a 0.7 megapixel image resized to 1.5.

That was my first reaction, too. I was disappointed. Every pixel could have been a 3x3 supersampled "superpixel" providing the best HD footage ever. We hoped for.

But then I realized that I was disappointed mainly because hopes die last.

There obviously are some serious technical obstacles to get this and Pentax engineers must have decided that it was good enough for a "goodie" (i.e., a bonus which didn't cost extra).

To get a better feeling what to realistically expect, I inspected some sample frames at 400% 1920x1080 from the "Canon HF10", the newest and seemingly best HDV camera in K20D's class of price and weight. The particular sample video was received very enthusiastically by the corresponding forum members.

Comparing head to head, I would say that both provide the same quality of frame IQ if the subject doesn't move and is in bright light. The K20D has somewhat better vertical resolution and somewhat worse horizontal resolution. Compared at 100% crop it is hard to see a difference in IQ. Assuming that K20D's images are treated as I described above to compensate for artifacts.

At bad light conditions, I guess that the K20D will win: better ISO performance and better glass. BTW, there are is no real wide angle for those portable HDV cameras.

So, what Pentax gave us is HDV quality for 117 frames. I know we hoped better but it could have been worse. And as somebody may have stated: the K20D is no movie camera

Nevertheless, I find it important to note that burst mode images should be PP'ed as decribed before. I mean, before selling this soccer goal shot or what else

raz · 03-03-2008, 03:18 PM

No this time I really don't agree with you

I have the Canon HV20, it has a CMOS sensor which is capable of capturing 1920x1080 video at 25p (25 progresive frames/second) and puts them on the tape at 1440x1080 (hdv format).

You cannot compare a photo camera that captures only 768x1024 pixels with a HD camcorder..even if the smaller image is somehow sharper. The true HD camera captures more details, at true pal framerate, it has functioning autofocus during the capture (and a very fast one), it has functioning auto exposure during the capture, it can record clear and crisp sound, it has a 10x zoom, and more of 5 seconds of recording.

I don't think the intention of Pentax was to compete or to say, look the camera is more capable than a HD camcorder. Also the resolution is to low, at 768x1024 pixels...I could say, look my K100D can output 3000x2000 pixel images, so if I resize the image to 4500x3000 and I gave it a little sharpen, then it will have "almost" the same detail compared to a K20D...so what's the point in having the K20D in the first place ?

falconeye · 03-03-2008, 03:27 PM

Originally posted by raz

No this time I really don't agree with you

This time I fully agree with you

In no way I wanted to say the K20D should be compared to a video camera.

I only compared 400% crops of both to get a feeling about each others 'state of the art' for single frame image quality (IQ). And at that point, I only say both are close, aren't perfect and capture about the same level of detail. The famous "true 1920x1080" from an HDV camera aren't crisp either (in the latter case it may actually be the lens' limit with 4 micron sensor cells and a 10x zoom).

But of course, they do not compete anyway.

Jonson PL · 04-20-2008, 03:04 AM

Thanks for your work, much appreciated

jsherman999 · 04-20-2008, 07:17 AM

Originally posted by raz

No this time I really don't agree with you

I have the Canon HV20, it has a CMOS sensor which is capable of capturing 1920x1080 video at 25p (25 progresive frames/second) and puts them on the tape at 1440x1080 (hdv format).

You cannot compare a photo camera that captures only 768x1024 pixels with a HD camcorder..even if the smaller image is somehow sharper.

All true, but there's something else to keep in mind here - the quality of the glass. The
HV20 (and other actual HD video cameras) are built for 25fps+ video capture - the Pentax
K20D burst mode is built for single frame capture from a set - not to make little movies, really,
although that's kind of a side effect available.

With the video camera, you're suck with the glass they give you. With the Pentax, you have
the option of changing lenses, to get different effects, resolution, better quality at extreme
telephoto. Again, this is meant for single frame capture from a set - it should be compared
more accurately to 6fps DSLR type shooting than full video, although it's kind of a hybrid
category.

Here's a snapshot taken at 21fps burst mode with the 77ltd - no PP was done, and I
haven't done any of the anti-aliasing falconeye talks about, which would quickly improve
the image even more. I think this shows a bit of the 77 magic, but if I were to work at it
a bit, I could get more - and I'd like to see what my DA* 50-135 could do with burst mode...
I don't think a HD video camera's glass can offer that flexibility.

Link to more --> 21fps_burst_mode_77ltd

.

03-02-2008, 05:24 PM	#1
falconeye Pentaxian	K20D burst mode has 768x1024 internal resolution In thread Fireengine K20D movie in full HD I have posted a fireengine video in full resolution MP4/AVC. Discussion on a German board have led to the conclusion that the "native" resolution of images captured in burst mode is 768x1024, then scaled up horizontally to 1536x1024 (1.6 MPixel). Here are two 400% crops from the original images in the video: Smallest details seem to be 2pixels wide by 1pixel high. This resolution is still ok for 720p HD video but is not exactly what to expect from the burst mode specification as given by Pentax. For everybody considering the burst feature, e.g., for sports photography, base your computations on a usable anamorphic resolution of 768 wide by 1024 high. At this resolution, however, the image is sharp at 100% crop and outperforms so-called HD images from consumer HD cameras (which look bad at a 100% crop).

03-02-2008, 05:52 PM	#2
falconeye Pentaxian	PP can much improve results from burst feature I've done one more test as a logical response to the finding above: One can treat the images captured in burst mode as follows: (1) Resample to 768x1024 resolution with NO interpolation (2) Resample the result to 1536x1024 resolution with bicubic interpolation The result are much more naturally looking images, the alias artifacts are gone: I recommend everybody who wants to use the burst feature to treat the resulting images in this way. Sort of Day/Night difference. However, one artifact isn't removed that way: There is still 2 pixel wide purple fringe on the right side of dark areas of all vertical contrast borders, and green fringe on the left side. This is the same for all locations in the image (left and right of center), so I guess it cannot be the lens. There is no PF issue with horizontal contrast borders, though. Does anybody know a software which removes this kind of constant PF from an image? I guess it comes from some strange binning done to get the burst images... Images from burst mode, when treated both for alias artifacts (cf. above) and the constant PF, are of good enough quality to be useful. But they shouldn't be used w/o further PP.

03-02-2008, 06:15 PM	#4
falconeye Pentaxian	Originally posted by jsherman999 How does one "Resample the result to 1536x1024 resolution with bicubic interpolation"? Ok, I assume basic post processing experience here. In Photoshop CS3, you use Menu Item "Image -> Image Size ..." (translated back from German, so the exact naming may differ) type in 768 as new width (change units to pixels and disable proptional sizes first as you aren't changing the height) and use "nearest neighbor" as resampling algorithm. use the Menu Item a second time, but now type 1536 as new width and use "bicubic" as resampling algorithm. I am sure, it is even simpler in Lightroom and can be scripted. And it should always be done when importing bursts as my sample images above may have illustrated.

Thread Tools	Search this Thread
Show Printable Version E-mail to a Friend	Search this Thread: Advanced Search

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Example of K20d burst mode	johnmflores	Post your photos!	5	05-21-2009 07:33 PM
K20D Burst Mode - Batter up	daacon	Post your photos!	10	05-04-2009 10:08 AM
K20D burst mode - post your vids!	hamidlmt	Pentax DSLR Discussion	11	09-17-2008 12:03 PM
K20D burst mode video	hamidlmt	Post your photos!	1	09-17-2008 10:40 AM
Fun with K20D, burst mode & figure-skating	nils	Post your photos!	10	03-24-2008 03:12 PM

03-02-2008, 05:58 PM	#3
jsherman999 Pentaxian	Originally posted by falconeye I've done one more test as a logical response to the finding above: One can treat the images captured in burst mode as follows: (1) Reample to 768x1024 resolution with NO interpolation (2) Reample the result to 1536x1024 resolution with bicubic interpolation The result are much more naturally looking images, the alias artifacts are gone. I recommend everybody who wants to use the burst feature to treat the resulting images in this way. Sort of Day/Night difference. How does one "Reample the result to 1536x1024 resolution with bicubic interpolation"? Thanks very much for your analysis, BTW. .

03-02-2008, 07:08 PM	#5
raz Senior Member	I doubt that what you are saying is right, because a 100% increase of resolution from 0.7 Mpixels to 1.5 would be to visible. Can you attach some shots (JPG directly from camera not mp4) of a burst ? I've seen some from others but I canot find the links anymore.

03-02-2008, 07:24 PM	#6
falconeye Pentaxian	Originally posted by raz I doubt that what you are saying is right, because a 100% increase of resolution from 0.7 Mpixels to 1.5 would be to visible. Seeing is believing. Just look at my example above. The simple point is that the 1.6 MPixel images from the Pentax burst mode are already upsampled, but badly so. So, downsampling to recover the internal version and doing it right improves much. Still no 1.6 MPixel sharpness, but now w/o the artifacts. And of course, one must upsample to recover the correct aspect ratio. Originally posted by raz Can you attach some shots (JPG directly from camera not mp4) of a burst ? There is nothing wrong with a frame from a good MP4/AVC. But for added convenience, here is one unprocessed original:[IMGWIDELEFT]http://www.livis.de/download/pub/fireengine.jpg[/IMGWIDELEFT]

03-02-2008, 09:33 PM	#7
raz Senior Member	Is this directly from the camera ? What settings did you use for jpgs ? This look very bad, the antialiasing is bad (it may be as you sugest a problem with the resampling...) I've seen something like this in the jpg of a preproduction model, but on dpreview forum on the ones with firmeware 1.0 I didn't noticed that problem ?

03-03-2008, 05:35 AM	#8
falconeye Pentaxian	Originally posted by raz on dpreview forum on the ones with firmeware 1.0 I didn't noticed that problem ? Unfortunately, you don't provide a source. I could only find one sample on dpreview with 1.00 firmware. It shows a squirrel. It also has this purple/green fringe problem, hence I believe it is taken from a K20D burst. However, the image is soft overall (so no alias artifacts to expect anyway) and moreover, was scaled down 83% at least. It may be that in this thread here, you see the first serious investigation of K20D burst mode resolution. Originally posted by raz Is this directly from the camera ? What settings did you use for jpgs ? This look very bad, the antialiasing is bad (it may be as you sugest a problem with the resampling...) As I have said before, JPEG directly from the camera (unlike the squirrel on dpreview), my JPEG setting was *** (but I guess it would be ignored anyway). Firmware is 1.00, otherwise you are invited to inspect the EXIF of my sample provided above. And yes, just trust me. It is exactly as "I sugest". I've invested enough time and have the background to be confident to say so. But it isn't this bad. As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?). What remains just is this thread's topic title: The K20D's internal burst video resolution is 768 x 1024.

03-03-2008, 09:30 AM	#9
jsherman999 Pentaxian	Originally posted by falconeye Unfortunately, you don't provide a source. I could only find one sample on dpreview with 1.00 firmware. It shows a squirrel. It also has this purple/green fringe problem, hence I believe it is taken from a K20D burst. However, the image is soft overall (so no alias artifacts to expect anyway) and moreover, was scaled down 83% at least. It may be that in this thread here, you see the first serious investigation of K20D burst mode resolution. As I have said before, JPEG directly from the camera (unlike the squirrel on dpreview), my JPEG setting was *** (but I guess it would be ignored anyway). Firmware is 1.00, otherwise you are invited to inspect the EXIF of my sample provided above. And yes, just trust me. It is exactly as "I sugest". I've invested enough time and have the background to be confident to say so. But it isn't this bad. As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?). What remains just is this thread's topic title: The K20D's internal burst video resolution is 768 x 1024. I concur with falconeye's findings. .

03-03-2008, 09:31 AM	#10
raz Senior Member	Originally posted by falconeye But it isn't this bad. As I have pointed out, the alias artifacts can be reduced by simple PP and the purple/green fringe can be removed as well (any suggestions on how to do this exactly, please?). Yes it is. You get half the resolution, or megapixels from what pentax is saying. Is a huge difference. I thought the 1536x1024 would be a super crisp image taken with full resolution and downsampled or at least taken with 9 super pixels - a group of (3x3). That had a lot of potential, not like a 0.7 megapixel image resized to 1.5.

03-03-2008, 02:46 PM	#11
falconeye Pentaxian	Originally posted by raz Yes it is. You get half the resolution, or megapixels from what pentax is saying. Is a huge difference. I thought the 1536x1024 would be a super crisp image taken with full resolution and downsampled or at least taken with 9 super pixels - a group of (3x3). That had a lot of potential, not like a 0.7 megapixel image resized to 1.5. That was my first reaction, too. I was disappointed. Every pixel could have been a 3x3 supersampled "superpixel" providing the best HD footage ever. We hoped for. But then I realized that I was disappointed mainly because hopes die last. There obviously are some serious technical obstacles to get this and Pentax engineers must have decided that it was good enough for a "goodie" (i.e., a bonus which didn't cost extra). To get a better feeling what to realistically expect, I inspected some sample frames at 400% 1920x1080 from the "Canon HF10", the newest and seemingly best HDV camera in K20D's class of price and weight. The particular sample video was received very enthusiastically by the corresponding forum members. Comparing head to head, I would say that both provide the same quality of frame IQ if the subject doesn't move and is in bright light. The K20D has somewhat better vertical resolution and somewhat worse horizontal resolution. Compared at 100% crop it is hard to see a difference in IQ. Assuming that K20D's images are treated as I described above to compensate for artifacts. At bad light conditions, I guess that the K20D will win: better ISO performance and better glass. BTW, there are is no real wide angle for those portable HDV cameras. So, what Pentax gave us is HDV quality for 117 frames. I know we hoped better but it could have been worse. And as somebody may have stated: the K20D is no movie camera Nevertheless, I find it important to note that burst mode images should be PP'ed as decribed before. I mean, before selling this soccer goal shot or what else

03-03-2008, 03:18 PM	#12
raz Senior Member	No this time I really don't agree with you I have the Canon HV20, it has a CMOS sensor which is capable of capturing 1920x1080 video at 25p (25 progresive frames/second) and puts them on the tape at 1440x1080 (hdv format). You cannot compare a photo camera that captures only 768x1024 pixels with a HD camcorder..even if the smaller image is somehow sharper. The true HD camera captures more details, at true pal framerate, it has functioning autofocus during the capture (and a very fast one), it has functioning auto exposure during the capture, it can record clear and crisp sound, it has a 10x zoom, and more of 5 seconds of recording. I don't think the intention of Pentax was to compete or to say, look the camera is more capable than a HD camcorder. Also the resolution is to low, at 768x1024 pixels...I could say, look my K100D can output 3000x2000 pixel images, so if I resize the image to 4500x3000 and I gave it a little sharpen, then it will have "almost" the same detail compared to a K20D...so what's the point in having the K20D in the first place ?

03-03-2008, 03:27 PM	#13
falconeye Pentaxian	Originally posted by raz No this time I really don't agree with you This time I fully agree with you In no way I wanted to say the K20D should be compared to a video camera. I only compared 400% crops of both to get a feeling about each others 'state of the art' for single frame image quality (IQ). And at that point, I only say both are close, aren't perfect and capture about the same level of detail. The famous "true 1920x1080" from an HDV camera aren't crisp either (in the latter case it may actually be the lens' limit with 4 micron sensor cells and a 10x zoom). But of course, they do not compete anyway.

04-20-2008, 03:04 AM	#14
Jonson PL Veteran Member	Thanks for your work, much appreciated

04-20-2008, 07:17 AM	#15
jsherman999 Pentaxian	Originally posted by raz No this time I really don't agree with you I have the Canon HV20, it has a CMOS sensor which is capable of capturing 1920x1080 video at 25p (25 progresive frames/second) and puts them on the tape at 1440x1080 (hdv format). You cannot compare a photo camera that captures only 768x1024 pixels with a HD camcorder..even if the smaller image is somehow sharper. All true, but there's something else to keep in mind here - the quality of the glass. The HV20 (and other actual HD video cameras) are built for 25fps+ video capture - the Pentax K20D burst mode is built for single frame capture from a set - not to make little movies, really, although that's kind of a side effect available. With the video camera, you're suck with the glass they give you. With the Pentax, you have the option of changing lenses, to get different effects, resolution, better quality at extreme telephoto. Again, this is meant for single frame capture from a set - it should be compared more accurately to 6fps DSLR type shooting than full video, although it's kind of a hybrid category. Here's a snapshot taken at 21fps burst mode with the 77ltd - no PP was done, and I haven't done any of the anti-aliasing falconeye talks about, which would quickly improve the image even more. I think this shows a bit of the 77 magic, but if I were to work at it a bit, I could get more - and I'd like to see what my DA* 50-135 could do with burst mode... I don't think a HD video camera's glass can offer that flexibility. Link to more --> 21fps_burst_mode_77ltd .