Go Back   Steve's Digicams Forums > Digicam Help > General Discussion

Reply
 
Thread Tools Search this Thread
Old Mar 28, 2008, 4:49 PM   #11
Junior Member
 
Join Date: Mar 2008
Posts: 4
Default

In reply to AC's point #4. It seems like you are saying that EIS is bad news in general. I have seen some pretty good implementations, but I can see that it still needs work. However, I will argue that it can certainly do better in the future, and that it doesn't have to reduce image quality in principle.



The most simple thing you can do is crop in x and y to reduce the effect of pan and tilt shake, and this doesn't reduce the quality, since you are just cropping.

If you get fancy, you can resample the frames to do a better job of stabilization. Resampling usually hurts image quality. However, the resample can also be used to remove some image warping from things such as the optics or CMOS rolling shutter warp.

However, if you get really fancy, you can use multiple aligned frames to reconstruct a higher quality image at each frame. You can get higher resolution (if the optics pass enough) and lower noise. This can even work in a still camera or in a playback device. Multiple aligned shots are routinely used to improve quality in astronomy. See "lucky imaging" and "speckle imaging" as some very tricky ways of combining multiple shots.

It is easy to make a bad IS system that just makes things look worse, and many people have. But there isn't anything inheirently bad about the idea. If, by electronic image stabilization, you mean:

Take all the frames that were captured and computationally reconstruct a sequence that uses all of the available information to make the smoothest, highest image quality video sequence that you can.

Then, clearly I think, it has the potential to make video better. If you use IE to mean:

Estimate camera motion (badly), then crop and resample to get approximatly the same image that a stable camera would have seen on that individual frame.

Then, yes, it will probably do some yucky things.

No camera takes all of the frames in a shot and does the first type of thing, but a good EIS system uses some information across several shots, so it can do a pretty good job. It isn't a mature technology yet, and as buffer sizes and computational power in the devices gets bigger, expect better things out of it.

On the other topic, pixel sizes are about as small as they can reasonably go, and don't expect anything good out of more, smaller pixels. The smallest pixels on the market, at 1.4 microns, are only twice the wavelength of red light across. Their efficiency falls off of a cliff as they approach the scale of the light they are supposed to be sensing. Also, the resolution of a system is limited by the F# of the lens due to diffraction. With a reasonably fast lens, the sharpest point that can be imaged is around 2 microns across. So going to smaller pixels isn't going to help resolution.

Sensor makers have done an amazing job of improving sensitivity while reducing size, but the improvements in image quality are not due to having small pixel size. For example, if they took what they have learned to make this year's 10MP still camera work, and went back and applied the same things to the 8MP-sized pixels, the 8MP camera would make better looking pictures than the 10MP camera. Image quality has not improved because they went to super high pixel counts. It has improved despite going to high pixel counts.

Having a bigger system, with big glass and a big chunk of silicon, with lots of big pixels, gets around this problem. (this is sort of what Gerrydsjr was asking about big lenses) So, yes, more is more, if you go big. But smaller, cheaper, and more, isn't going to work if it requires the individual pixels to get smaller.

Regards,

Amnon

Amnon is offline   Reply With Quote
Old Mar 28, 2008, 8:38 PM   #12
Senior Member
 
BillDrew's Avatar
 
Join Date: Jun 2002
Location: Hay River Township, WI
Posts: 2,512
Default

Amnon wrote:
Quote:
... and this doesn't reduce the quality, since you are just cropping.
...
When cropping is done properly, it will often improve the quality of an image. However it will always reduce the resolution as measured by pixel count. This is another example of pixel count not being a complete measure of quality.
BillDrew is offline   Reply With Quote
Old Mar 29, 2008, 9:51 PM   #13
Senior Member
 
Join Date: Dec 2006
Location: Davenport, IA
Posts: 2,093
Default

Amnon wrote:
Quote:
1) In reply to AC's point #4. It seems like you are saying that EIS is bad news in general. I have seen some pretty good implementations, but I can see that it still needs work. However, I will argue that it can certainly do better in the future, and that it doesn't have to reduce image quality in principle.
Quote:
2) Sensor makers have done an amazing job of improving sensitivity while reducing size, but the improvements in image quality are not due to having small pixel size. For example, if they took what they have learned to make this year's 10MP still camera work, and went back and applied the same things to the 8MP-sized pixels, the 8MP camera would make better looking pictures than the 10MP camera. Image quality has not improved because they went to super high pixel counts. It has improved despite going to high pixel counts.

Regards,

Amnon
Comment 1: The correct interpretation of what I'm saying is that good implementation seem to have been accomplished for video systems but the evidence is that it isnot useful at this point for still photos. Optical stabilization clearly is useful for still photography and I have heard no ill words about it in video.

Comment 2: It wasn't the fact that thepixel was smaller that made the resolution improvement in the S5 but rather the greater number of pixels. If evolutionary improvements in sensor technology permit the creation of a 10MP 1/2.5 sensor with the samepixel S/N ratio as last year's 8MP 1/2.5 sensor there is absolutely no incentive to roll that technology back into a 8MP 1/2.5 sensor. That' reality. There might be incentive to apply that technology to a 15MP APS-C sensor however.

ac.smith is offline   Reply With Quote
Old Apr 2, 2008, 1:39 PM   #14
Junior Member
 
Join Date: Mar 2008
Posts: 4
Default

Regarding AC Comment 1

I think the good versions of still image stabilization are just starting to come out now. Wait and you'll see.



Regarding AC Comment 2

There is a big advantage of stepping back a pixel size! Consider 2.2, 1.75 and 1.4 micron pixels. This is the current trend in cellphone camera chips. Last year's cellphones with 2.2 micron pixels are just barely usable in low light. The noise was terrible. With a lot of work, you can get the same terrible performance at 1.75 microns, so that is what they did for this year's crop. Use every trick in the book, and you can get equally awful performance at 1.4 microns, and that is the plan for next year. Now, keeping in mind that the spot size of a diffraction-limited cellphone camera lens is larger than 2 microns on axis, what havewe gained with these advances? Pretty much nothing. The true system resolution is no better. Now, if you backed off to 2.2 microns, the resolution would be almost the same (and I am using resolution in the sense of resolvable detail, not pixel count), but your signal to noise ratio would be much much better. Instead of getting a high megapixel number, super-sampled blur with tons of noise, you would instead get a well-sampled image with very low noise.

I would much rather get big ISO numbers with low noise than big megapixel numbers with high noise, wouldn't you?

In devices with tiny pixels the argument is easy to make, since the pixel size is getting smaller than the lens can resolve. However, I will argue, the same thing goes for devices with big pixels and big lenses, such as DSLRs with 6 micron pixels. Since the fill-factor (the ratio of active sensor area relative to the total pixel sixe) gets much better with bigger pixels, there is a lot to be said for making the pixels somewhat larger, instead of pushing the size smaller and smaller every year. We keep getting cameras with about the same amount of noise (e.g. a fair-quality ISO 800 image) and bigger megapixel numbers. I would be happier if I could get ISO 3200 with low noise and a modest megapixel number, like say 6MP, instead of an ISO 800 12MP camera. Of course, it depends on what you want to do with your photos, what sort of photos you shoot, if you like making big prints, how much noise vs. sharpness you like and etc. So this argument isn't as clear-cut as the cellphone example. However, since consumers only seem to notice the megapixel number, I think the megapixel number has been pushed at the cost of all else. So in general, megapixel numbers have gone too far and most people would be happier with lower megapixel cameras with better ISO numbers.

Shooting with a high-ISO camera and a fast lens is a real joy. You have so many more options. You can stop down, you can shoot short exposures, you can shoot without a flash, etc. You can just do more. If you can push the ISO numbers up, you can get the same thing out of a cheaper lens, and do even more with a faster lens.

If your camera had a HIGH ISO + LOW MEGAPIXEL <-> LOW ISO + HIGH MEGAPIXEL knob on it, and you could magically adjust this parameter, I bet I would find that knob turned way over to the high ISO andlow megapixel side more often than not.

For the higher-end devices, the manufacturers have pegged this knob to the high megapixel end.And for the low-end consumer imaging devices (e.g. cellphone cameras) the manufacturers have maladjusted this knob past the end, to the horrible ISO + absurd megapixel level.




Amnon is offline   Reply With Quote
Old Apr 2, 2008, 3:00 PM   #15
Senior Member
 
Join Date: Dec 2006
Location: Davenport, IA
Posts: 2,093
Default

Amnon wrote:
Quote:
Regarding AC Comment 1

I think the good versions of still image stabilization are just starting to come out now. Wait and you'll see.



Regarding AC Comment 2

There is a big advantage of stepping back a pixel size! Consider 2.2, 1.75 and 1.4 micron pixels. This is the current trend in cellphone camera chips. Last year's cellphones with 2.2 micron pixels are just barely usable in low light. The noise was terrible. With a lot of work, you can get the same terrible performance at 1.75 microns, so that is what they did for this year's crop. Use every trick in the book, and you can get equally awful performance at 1.4 microns, and that is the plan for next year. Now, keeping in mind that the spot size of a diffraction-limited cellphone camera lens is larger than 2 microns on axis, what havewe gained with these advances? Pretty much nothing. The true system resolution is no better. Now, if you backed off to 2.2 microns, the resolution would be almost the same (and I am using resolution in the sense of resolvable detail, not pixel count), but your signal to noise ratio would be much much better. Instead of getting a high megapixel number, super-sampled blur with tons of noise, you would instead get a well-sampled image with very low noise.

I would much rather get big ISO numbers with low noise than big megapixel numbers with high noise, wouldn't you?

In devices with tiny pixels the argument is easy to make, since the pixel size is getting smaller than the lens can resolve. However, I will argue, the same thing goes for devices with big pixels and big lenses, such as DSLRs with 6 micron pixels. Since the fill-factor (the ratio of active sensor area relative to the total pixel sixe) gets much better with bigger pixels, there is a lot to be said for making the pixels somewhat larger, instead of pushing the size smaller and smaller every year. We keep getting cameras with about the same amount of noise (e.g. a fair-quality ISO 800 image) and bigger megapixel numbers. I would be happier if I could get ISO 3200 with low noise and a modest megapixel number, like say 6MP, instead of an ISO 800 12MP camera. Of course, it depends on what you want to do with your photos, what sort of photos you shoot, if you like making big prints, how much noise vs. sharpness you like and etc. So this argument isn't as clear-cut as the cellphone example. However, since consumers only seem to notice the megapixel number, I think the megapixel number has been pushed at the cost of all else. So in general, megapixel numbers have gone too far and most people would be happier with lower megapixel cameras with better ISO numbers.

Shooting with a high-ISO camera and a fast lens is a real joy. You have so many more options. You can stop down, you can shoot short exposures, you can shoot without a flash, etc. You can just do more. If you can push the ISO numbers up, you can get the same thing out of a cheaper lens, and do even more with a faster lens.

If your camera had a HIGH ISO + LOW MEGAPIXEL <-> LOW ISO + HIGH MEGAPIXEL knob on it, and you could magically adjust this parameter, I bet I would find that knob turned way over to the high ISO andlow megapixel side more often than not.

For the higher-end devices, the manufacturers have pegged this knob to the high megapixel end.And for the low-end consumer imaging devices (e.g. cellphone cameras) the manufacturers have maladjusted this knob past the end, to the horrible ISO + absurd megapixel level.



CMT 1: I don't see that at all, I see ois being incorporated into more and more P&S cameras. Kodak has, just this year, incorporated it into some of their 5x and 3x lenses. Canon is steadily incorporating it into more of their P&S line at all price levels. Digital stab. in still cameras seems to be have been viewed as a stop gap measure by the manufacturers until they could apply an ois solution.

CMT2: I am not arguing that small pixels are better. I am saying that that the technology thatdelivers an equal or betterS/N ratio in this year's 8MP 1/2.5 sensor compared to last year's 6MP 1/2.5 sensor will not be rolled back into a new 6MP 1/2.5 sensor no matter how much you what it because that's not the dynamics of the marketplace.

Actually many cameras have a control suchas you mention because to access the highest ISOs available you do have to reduce the resolution. If you wish to argue that the knob's already been turned past the stop at that point I'm OK with that. Mostly an unused feature I think.
ac.smith is offline   Reply With Quote
Old Apr 2, 2008, 5:37 PM   #16
Junior Member
 
Join Date: Mar 2008
Posts: 4
Default

Regarding 1, I don't mean that there isn't any room for ois and that electronic will do everything for you. I just mean that electronic has its place, it will be better in the future, and it will be better than nothing. I bet some form of eis will be in almost everything. Optical will grow at the same time, and it won't hurt to have both.

Regarding 2, I think we are in agreement. Up until now, like you say, the SNR has more or less held as pixel size has shrunk. Further, going to smaller pixels has increased resolution in a real sense (not just the pixel count). But as we go forward, no gains in resolution will be had from smaller pixels and SNR will get worse. Especially as we hit this turning point, as thesensor technology matures, I think consumers will realize that there is a tradeoff between pixel size and SNR, and they will start putting more weight on SNR. I think consumers are already getting wise to the idea that megapixels don't equal quality.
Amnon is offline   Reply With Quote
Old Apr 2, 2008, 7:41 PM   #17
Senior Member
 
TCav's Avatar
 
Join Date: Sep 2005
Location: Washington, DC, Metro Area, Maryland
Posts: 13,543
Default

rinniethehun wrote:
Quote:
Let the flood gates open...
Nice call.

Predicting the future is easy. Being right is tough.
TCav is offline   Reply With Quote
Old Apr 2, 2008, 7:52 PM   #18
Senior Member
 
TCav's Avatar
 
Join Date: Sep 2005
Location: Washington, DC, Metro Area, Maryland
Posts: 13,543
Default

Amnon wrote:
Quote:
...megapixels don't equal quality.
AT HIGH ISO SETTINGS!!!

The signal to noise ratios get bad when the signal is amplified by cranking up the ISO setting. Shoot at ISO 100, or even 400 or maybe even 800 (with an APS-C or larger image sensor) and there's no visible noise, so you get more megapixels (higher resolution)with no downside.

So if you can keep the ISO setting down, megapixels equal quality! (... all other things being equal.)
TCav is offline   Reply With Quote
Old Apr 2, 2008, 8:43 PM   #19
Senior Member
 
Join Date: Mar 2005
Posts: 1,870
Default

I thought it would be worse...

the Hun

rinniethehun is offline   Reply With Quote
Old Apr 3, 2008, 7:16 AM   #20
Senior Member
 
TCav's Avatar
 
Join Date: Sep 2005
Location: Washington, DC, Metro Area, Maryland
Posts: 13,543
Default

TCav wrote:
Quote:
Amnon wrote:
Quote:
...megapixels don't equal quality.
AT HIGH ISO SETTINGS!!!

The signal to noise ratios get bad when the signal is amplified by cranking up the ISO setting. Shoot at ISO 100, or even 400 or maybe even 800 (with an APS-C or larger image sensor) and there's no visible noise, so you get more megapixels (higher resolution)with no downside.

So if you can keep the ISO setting down, megapixels equal quality! (... all other things being equal.)
Which raises another issue with a popular form of digital image stabilization, the form that increases the shutter speed while increasing the ISO setting.

The result:
  1. Less motion blur due to camera shake. [/*]
  2. Correct exposure, because as the faster shutter speed decreases the EV, the higher ISO setting increases the EV a corresponding amount. [/*]
  3. Noise resulting from the higher ISO setting.[/*]
TCav is offline   Reply With Quote
 
Reply


Thread Tools Search this Thread
Search this Thread:

Advanced Search

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off



All times are GMT -5. The time now is 10:28 PM.