Deep depth of field to get maximum detail on human faces

[Eildosa]

Frankly, I rather doubt that the Canon will do anything a lot differently than the other cameras you have been working with. These low-end point & shoot cameras have pretty well equal performance optically, regardless of the brand badge on them.

I find it interesting that the Disney team (in the link you sent out) is using four high end cameras for their work, rather than a fleet of small, cheap ones. I suspect you might get identical results, with a lower cash outlay if you shot with run-of-the-mill webcams.

http://www.cameralabs.com/reviews/Ca...werShot_A2500/ no surprises in the review...

Actually if you read carefully, at some point they show some results from some point and shoot cameras, the result as almost as good as the DSLR one.
Because of the mesotropric algorythm.

The main problem in my setup is that there is only an "auto mode" in my coolpix L27, each camera has a different ISO and shutter speed setting, reducing the scan quality.

[Eildosa]

Frankly, I rather doubt that the Canon will do anything a lot differently than the other cameras you have been working with. These low-end point & shoot cameras have pretty well equal performance optically, regardless of the brand badge on them.

I find it interesting that the Disney team (in the link you sent out) is using four high end cameras for their work, rather than a fleet of small, cheap ones. I suspect you might get identical results, with a lower cash outlay if you shot with run-of-the-mill webcams.

http://www.cameralabs.com/reviews/Ca...werShot_A2500/ no surprises in the review...

Look at page one figure 1, on the left reconstruction using DSLR, on the right reconstruction using consumer cameras.

Right now, with 21 coolpix L27, with only auto mode (some picture are at ISO 100 some at 1000, some of them were taken with a shutter speed of 1/60 some at 1/10)
I'm getting unprecise scan, but they are not so bad either.

The eyes are not detailed enought, I get some bump here and there.
But keep in mind that this is done with a non specialized algorithm, the point matching stage could be enhance by detecting face feature. Plus I have no Mesoscopic augmentation like disney reseach.

It's true I won't have the same quality as with DSLR.

But I'm confident that being able to use the same ISO, aperture opening and shutter speed for all my camera will greatly enhance the result.

[Saorsa]

The Nikon Coolpix P series point and shoots do provide a lot of control over the camera as opposed to the L and S series.

You can probably find used Nikon P300s at a reasonable price. I don't know which of the Canon or Panasonic models would suit your needs but the 'L'ife and 'S'tyle models aren't likely what you need in the Nikon line.

[Eildosa]

The Nikon Coolpix P series point and shoots do provide a lot of control over the camera as opposed to the L and S series.

You can probably find used Nikon P300s at a reasonable price. I don't know which of the Canon or Panasonic models would suit your needs but the 'L'ife and 'S'tyle models aren't likely what you need in the Nikon line.

Yes but the canon can load custom firmware from the SD card, allowing you to do ANYTHIIIIING.
Like shooting in raw with a 50 euro camera.
Or make you own communication protocol and speak with all cameras from usb

Frankly, I rather doubt that the Canon will do anything a lot differently than the other cameras you have been working with.

Problably true for the image quality. however I'll be able to force all the camera to shoot with the same settings, scan quality will be greatly enhanced.

Disney research did use consumer camera, look at figure 1.
I don't really see a difference in quality from the DSLR and the compact camera (I'm talking about the 3d model, not the texture which will obviously be better with DSLR)

Furthermore I don't need 21 cameras for the face, (10 are sufficient, even 8).
At first this was a full head setup, but because some cameras select ISO 1000 or 2000 half of the pictures cannot be used.
So I moved all the camera to the face to be able to always get enought exploitable pictures.

[chauncey]

Aah, why are you using cameras...would not laser scanners be more effective in meeting your goals?

[Eildosa]

Aah, why are you using cameras...would not laser scanners be more effective in meeting your goals?

Laser scan takes a lot of time to acquire the data, a camera scan last 1/25 seconds in the worst case scenario
A laser scan more than 10 second.

I don't know anybody able to hold an emotion without moving AT ALL for that long.

[Eildosa]

Here are the many ways you can control the depth of field:

http://media.digitalcameraworld.com/...heat_sheet.jpg

I did not see that!
Well, then putting the camera 2 meters behing is useless if I use the Zoom to get the face to fill the whole pictures.

If I understand correctly, smaller aperture = more DOF
And with the same aperture, a subject placed at 2m will get a shallower DOF than one a 5 meters.

now, with an aperture set to F8 and the subject being at 5 meter, I should get a wide enought DOF in order to get all it's face in forcus.
Only if I do not Zoom on it's face in order to get it to fill the whole picture.
Am I right?
Which is kind of bad because the less pixel about the face I get the less detail I'll get during reconstruction.

[chauncey]

Now that I understand your goals...why not use a 4-K video camera and lift stills from the video recording?
Put your subject on a motorized turntable and record while he spins...for different
perspectives, more than one recorder might be needed using that same arc arrangement.
You should be able to rent them to save budget.

[Manfred M]

I did not see that!
Well, then putting the camera 2 meters behing is useless if I use the Zoom to get the face to fill the whole pictures.

If I understand correctly, smaller aperture = more DOF
And with the same aperture, a subject placed at 2m will get a shallower DOF than one a 5 meters.

now, with an aperture set to F8 and the subject being at 5 meter, I should get a wide enought DOF in order to get all it's face in forcus.
Only if I do not Zoom on it's face in order to get it to fill the whole picture.
Am I right?
Which is kind of bad because the less pixel about the face I get the less detail I'll get during reconstruction.

Again, yes but it is not just about DoF, but also more accurate capture for the software to overlay. A longer shot (which by the way, common practice in portrait photography) gives you a flatter (more natural) rendering, whereas a shorter focal length will introduce distortion. I would imagine that the image conversion software would have more useable data that way.

The second issue is lens characteristics; the mid-range of most lower end lenses exhibit less distortion than the extremes (this is especially true about the wider angle settings on a low end camera like you are using). Again, better datapoints from image to image for the software to work with = a better end product due to better input data.

DoF is only one consideration.

[Manfred M]

Now that I understand your goals...why not use a 4-K video camera and lift stills from the video recording?
Put your subject on a motorized turntable and record while he spins...for different
perspectives, more than one recorder might be needed using that same arc arrangement.
You should be able to rent them to save budget.

Read the previous threads; he shoots live people so a single camera technique isn't going to work for him.

Personally, I would probably cluster about 8+ low-end manual Chinese flashes together in the setup, behind some heavy duty diffusers instead of continuous lights. This would provide a higher shutter speed and less blurring in the data capture operation.

[chauncey]

a single camera technique isn't going to work for him

Can't argue with that, which is why I suggested using multiple 4K recorders. Maybe five of them as the OP has used with the individual still cameras...rented of course, then lifting appropriate stills from the video.

[juznobsrvr]

Can't argue with that, which is why I suggested using multiple 4K recorders. Maybe five of them as the OP has used with the individual still cameras...rented of course, then lifting appropriate stills from the video.

I will have to try that one. I'm getting tired of taking taking two pictures, each one with focused to different eye, and compositing them. Haha.

[chauncey]

Just remember that the better ones cough out 24 frames per second shooting in RAW...that is a lot of
still images to filter through.
Another thought that just occurred to me would be to create a mask...https://www.youtube.com/watch?v=E5HeKECvnmU

[Manfred M]

Just remember that the better ones cough out 24 frames per second shooting in RAW...that is a lot of
still images to filter through.
Another thought that just occurred to me would be to create a mask...https://www.youtube.com/watch?v=E5HeKECvnmU

I assume you are thinking of something like the Panny Lumix GH4? $1800 + lens. I wonder if the Lumix 3-D lens might do it in one shot. It's a fixed 12.5mm, but at f/12; pretty darn slow.

[ajohnw]

It seems to me that as is often the case the OP needs a better understanding of depth of field. There was an informative thread on this subject recently here

Depth of field 2/3 behind and 1/3 in front???

But a few things to note before reading it.

If the OP frames a head the same aperture will give the same depth of field irrespective of the focal length used.

A lens is only at it's sharpest at the distance it's focused to and deteriorates each side of this point. How apparent that deterioration is depends on how large the image is when viewed.

It's generally not a good idea to view images taken with this type of camera full sized. All sorts of strange things go on inside of them to try and improve the quality of the image they produce. They are mainly aimed at producing smaller images for viewing on a PC screen especially when there is a lot of fine detail in them. Neither the sensor or the lenses are really up to it.

The results posted initially at the 2 different apertures could vary for a number of reasons. Maybe the camera didn't focus at exactly the same point. Maybe the software in the camera couldn't reconstruct the detail well in jpg form from the slower aperture - with sensors and pixels of this size diffraction and lens defect effects will be more apparent. Looking at the full sized image there is only a marginal difference anyway. I wouldn't call either sharp.

This is around the maximum size of final image that you can really expect for shots like this. Notice that the detail is barely visible. I sharpened your softer image.



This is the full sized version of that

http://filebin.net/yxpupz6pwj

This is a reduced image that has been tone mapped to increase contrast in low contrast areas. Detail is a little more visible.



Post processing might be an answer to your problem but looking at the example you posted this sort of detail level isn't visible anyway so you are probably expecting too much from the process. The Disney examples look augmented to me as well and only have 1280 pixel resolution full sized. That corresponds to a 1.6mp camera. If you reduce your images to this size I suspect you will get better results but retaining fine detail such as in the heads eyes is probably not possible.

John
-

[chauncey]

I'm still thinking that there has to be an easier way...enter YouTube...
https://www.youtube.com/results?sear...year&lclk=year

[Eildosa]

If the OP frames a head the same aperture will give the same depth of field irrespective of the focal length used.

this contradict the third example of this infographic (I'm guessing that each one is framing it's subject the same way): http://media.digitalcameraworld.com/...heat_sheet.jpg

Post processing might be an answer to your problem

that would be too easy
Sharpening is forbidden, it introduce noise, the noise pattern being different on each sharpened pictures it creates "snow" around the reconstruction and the reconstruction will look like it's made of sand.
As a matter of fact ANY post processing is forbidden.
And yes your camera does a bit of post processing, but I can't do anything about that beside not adding an other layer of post processing over it.

only have 1280 pixel resolution full sized

1280x1280 is not the resolution of the pictures but the area in the pyramidal matching process :
http://image.ntua.gr/iva/files/hpm_ijcv.pdf
Look at page 6.
In photogrammetry the higher the sharpness and the higher the resolution is better, reducing you pictures to 1280px while your camera gives you 4000+px pictures would be madness.

Again, I won't be able to do perfect shots like thos studio who uses 64* 1500$ DSLR.
I'm just trying to get the most perfect possible shot from my camera.

I'm still thinking that there has to be an easier way...enter YouTube...
https://www.youtube.com/results?sear...year&lclk=year

Sadly no,
If you just want a neutral head you could actually make a pretty good scan with only one DSLR.
But since I want to be able to do emotions the multi-cam setup is the only way to go.

It could be easier if I had a whole empty room.
I could put the cameras 3 meter away and use various light under white umbrellas.
Here since the wood support are very close if I but the light behing they will cast a shadow.

[chauncey]

I posed this poser to one of my geeky kids...he suggested this http://www.123dapp.com/catch

[Eildosa]

I posed this poser to one of my geeky kids...he suggested this http://www.123dapp.com/catch

I tried them all but thanks.

Agisoft photoscan
photomodeler
3D zephyr
123d catch.

123D catch is the one giving best result with ****ty pictures, In my first scan the white balance was in auto, some camera where wrongly setted and had different exposure setting.
agisoft would make a big potatoe while 123d catch did something not bad.

However if you have good quality pictures you can throw 123d catch to the trash.
123d catch always resize you pictures to 3Mp and thus cannot generate things with a very high level of details.

The problem is that they are "dumb" meaning they don't know what they are reconstructing.
The software I am developping will know that it has to deal with with a human face.

Because of this, point matching will be greatly enhence by first using face feature to orient pictures.
The generated 3D model will directly be exploitable by deforming a well made generic head to fit the result.
And I want to add the mesotropic augmentation from disney reseach on high resolution scan.

[ajohnw]

this contradict the third example of this infographic (I'm guessing that each one is framing it's subject the same way): http://media.digitalcameraworld.com/...heat_sheet.jpg


that would be too easy
Sharpening is forbidden, it introduce noise, the noise pattern being different on each sharpened pictures it creates "snow" around the reconstruction and the reconstruction will look like it's made of sand.
As a matter of fact ANY post processing is forbidden.
And yes your camera does a bit of post processing, but I can't do anything about that beside not adding an other layer of post processing over it.


1280x1280 is not the resolution of the pictures but the area in the pyramidal matching process :
http://image.ntua.gr/iva/files/hpm_ijcv.pdf
Look at page 6.
In photogrammetry the higher the sharpness and the higher the resolution is better, reducing you pictures to 1280px while your camera gives you 4000+px pictures would be madness.

Again, I won't be able to do perfect shots like thos studio who uses 64* 1500$ DSLR.
I'm just trying to get the most perfect possible shot from my camera.


Sadly no,
If you just want a neutral head you could actually make a pretty good scan with only one DSLR.
But since I want to be able to do emotions the multi-cam setup is the only way to go.

It could be easier if I had a whole empty room.
I could put the cameras 3 meter away and use various light under white umbrellas.
Here since the wood support are very close if I but the light behing they will cast a shadow.

Well I did point out that many need help getting to grips with DOF. Even the noddy page link that was posted. From an image quality point of view you will probably find that it's best to stick to shorter focal length settings. That is not an uncommon aspect of many zoom lenses especially as the price goes down.

Cameras. Looking at sample images in reviews from the Panasonic and the Canon you mention the in built noise reduction and sharpening in the Panasonic is noticeably better than the Canon. I'd also guess with good reason that the 100 ISO on both is actually higher and made to happen with software that plays with the numbers. This is even beginning to happen on some DSLR's. In short all sorts of things happen to the information from the sensor in a compact before it gets into the jpg image. That has allowed compacts to get were they are these days. Things were a lot better in this respect when they had around 6mp and 1/2 in sensors but that size of sensor rules out very long focal length zoom lenses.

There are lots of people including me on here that would love a problem free compact or even better a bridge camera with a long zoom but we are aware that even at the bridge camera level they are only really suitable for producing relatively small images in mostly ideal conditions. You have the ideal conditions but ......... need to cope with the rest.

John
-