Wolverine-Hawkeye Telecine

[Stan Jelavic]

Ok sounds like a plan to me Jarle.
Back to HDR. Spent a few days with Visual Basic .NET and compiled the imaging source samples and did some analysis on 12 bit images from the UX226.
If the image is underexposed the shadows turn black but if post-processed by adjusting gamma the shadows come out from the dark and look very good. The problem is that the bright areas get overexposed.
So I use the original image and the one with the gamma fix and ran them through the HDR script. The result turned out remarkably good. HDR from a single image.. pretty cool.
Here is one of the results.
https://photos.app.goo.gl/2u9bznjqnQydMrDf9




The gamma adjust was done in photoshop. Now I have to figure out how to do it in the VB app.
One more note - have not tried this with the 8 bit images but suspect that there will be some banding issues but not sure.

[Bruce Davis]

Hi Stan, would this produce a better hybrid image result than two exposures, or do two exposures still fare better overall. I would have thought that far more dark detail would be available with a longer exposure time, which of course blows out the lighter areas. I have read that this method is sometimes used in industrial applications but do not know how it compares with double exposure.

Had a play with gamma settings on the Capture_Low images, did not work out so well. By increasing the brightness in using gamma reveals a lot of noise and colour change, maybe its just me.


Thank you Stan
Regards - Bruce

[Mattias Norberg]

Hi Stan, would this produce a better hybrid image result than two exposures, or do two exposures still fare better overall. I would have thought that far more dark detail would be available with a longer exposure time, which of course blows out the lighter areas. I have read that this method is sometimes used in industrial applications but do not know how it compares with double exposure.

Had a play with gamma settings on the Capture_Low images, did not work out so well. By increasing the brightness in using gamma reveals a lot of noise and colour change, maybe its just me.


Thank you Stan
Regards - Bruce

if you are trying the .bat scrip i did post use default gamma setting on Stan´s image sequence he did share it works better
you can test to go up or down in small steps like 0.1 but the hdr picture i did post was default gamma = 1.0 on Low and High in POST Nr:#816 in the same post you have the .bat code without the imagemagick gamma change

SET Low_Exposure_Gamma=1.0
SET High_Exposure_Gamma=1.0

but on my own camera capture i have to have like this then Enfuse 4.2 start to work good

SET Low_Exposure_Gamma=2.0
SET High_Exposure_Gamma=3.0

[Stan Jelavic]

Hi Bruce and Mattias.
The method that I used that takes a single image, changes the gamma, and then combines with the original seems to work well with 16 bit images that I obtained from the UX226.
8 bit images may cause bands to appear and I believe that is what happened to you Bruce.
Additionally, the app that is used to change the gamma has to be in 16 bit mode.
My app works with 16 bits.
The histogram on the changed image is still nice a sooth. If you do it in 8 bit mode the histogram develops gaps and that causes the bands to appear.

I am still cleaning up the app and will be posting it here soon.

BTW - capture low images are 8-bit.
You can try it on this one:
https://drive.google.com/file/d/13QO...ew?usp=sharing
Make sure you app is in 16 bit mode.

[Mattias Norberg]

Hi Bruce and Mattias.
The method that I used that takes a single image, changes the gamma, and then combines with the original seems to work well with 16 bit images that I obtained from the UX226.
8 bit images may cause bands to appear and I believe that is what happened to you Bruce.
Additionally, the app that is used to change the gamma has to be in 16 bit mode.
My app works with 16 bits.
The histogram on the changed image is still nice a sooth. If you do it in 8 bit mode the histogram develops gaps and that causes the bands to appear.

I am still cleaning up the app and will be posting it here soon.

BTW - capture low images are 8-bit.
You can try it on this one:
https://drive.google.com/file/d/13QO...ew?usp=sharing
Make sure you app is in 16 bit mode.

i did test the hdr-lo.tiff 16bit in Photoshop but i did not get more detail from the shadows

i did test it with imagemagick it works with 16bit but i did test to pump up the gamma but did not get more detail from the shadows

on the right NO HDR picture you can see that there is more details in the shadows https://photos.google.com/share/AF1Q...9JM0J0TUFhc3NB

Stan have you lookt at OpenCv c++ API here is some tutorials https://www.opencv-srf.com/p/introduction.html
maybe it´s posible to add it to the IC Capture i do not know im not a Programmer
but if it works it can Align and do gamma and do HDR merge and much more

i been testing this totorials
https://www.learnopencv.com/image-al...encv-c-python/
https://docs.opencv.org/3.4/da/d5c/t..._detector.html
https://docs.opencv.org/3.4/d3/db7/t...r_imaging.html


and here is one that ask about super 8mm sprocket align OpenCV https://stackoverflow.com/questions/...s-using-opencv

[Bruce Davis]

Hi Stan and Mattias, copied Stan's "16bit hdr-lo.tiff" into photoshop to adjust the Gamma (no clipping occurs as you know), the sky remains the same and other areas of the image come up nicely, but dark areas not so much. Dark areas produce a lot of noise on the image sensor.

Using Photoshop HDR Toning, Gamma 0.70 and Exposure +0.05 :-


Just to illustrate/show the noise produced in the dark area on the right side of the image, just used the default settings in HDR Toning.

If two exposures are used hi & low, will the result be the same, will there be as much noise produced in the dark areas.
I used photoshop because it is easier for me as I do not understand anything concerned with programming.
Thanks and Regards - Bruce

[Stan Jelavic]

Bruce, I use gamma set to 0.5 and apply it to individual RGB components of the image. That is why the colors look washed out.


Combine that image with the original and get:
vs


I am running this on a video clip and will do the side by side, no-hdr and single image hdr. This particular scene has medium shadows
so the impact of HDR is not very apparent but I can see some improvements in the dark areas.
That dark window on the right hand side definitely has more detail.
When you zoom in you can see some noise caused by the camera. Possibly a better camera would make that better. Before we started doing this it did not matter that much because those areas were just very dark but now it is more visible.

Here is the single vs two image approach comparison side by side



Two image approach has somewhat less noise. Looks like it needs a bit more gamma to remove some fogginess.
Other than that I do not see much difference.

[Stan Jelavic]

Ok back to the drawing board. Ran a clip with the single image HDR and some scenes do come out OK and the sky is not as washed out but some scenes come out just too dark.
Mattias I used OpenCV for several years at work on a Linux platform for image processing and 3D cloud analysis and pretty familiar with it but have not used it for HDR. Could try it.

[Mattias Norberg]

Ok back to the drawing board. Ran a clip with the single image HDR and some scenes do come out OK and the sky is not as washed out but some scenes come out just too dark.
Mattias I used OpenCV for several years at work on a Linux platform for image processing and 3D cloud analysis and pretty familiar with it but have not used it for HDR. Could try it.

ok

i did see that the UX226 has Pixel size H: 1.85 µm, V: 1.85 µm that is very small so it must have low Dynamic range but maybe it have some new technics that it does not matter it have good Dynamic range anyway

i do know when i did buy my Camera about 7 years ago then it was better to have as big Pixel size as possible for very good Dynamic range in the picture my camera have Pixel size H: 4.4 µm, V: 4.4 µm

if i remember right someone on some forum did say that if the camera have Pixel size H: 6.0 µm, V: 6.0 µm or bigger then it can capture whole super 8mm Dynamic range then no need to do hdr

[David Brown]

Hello

Big shout out to Stan, Bruce and Mattias along with anyone contributing the HDR effort.

I just got the MSP430 flashed and it's working!!! For feedback on the V12_update (am I the only one?) The takeup reel is behaving very well. That is in RUN & RUN+TURBO. SPEED, while far better that before, has a little slack early. I think it could only be a problem for the Capstan. I don't have that physically installed right now. NICE.

Next, I will attempt to look at these HDR line commands and or HDR that I already have.
I have a PhotoMatix plugin for Lightroom that processes at 32bit. No banding there! Better even if the AVI is reduced back to 8bit when rendered. Lightroom might be too cumbersome to be practical.


Sorry to hear BACK TO THE DRAWING BOARD spoken. That's tough.

If it's easy to pad the numbered tifs, it would be great. I'm pretty sure I can renumber them anyway. They now sort as 1,10,100,101,102 etc.

The two memorable images in the recent posts for HDR were these:

post#797 by Mattias. "some examples i did not use davinci resolve for color correction only avisynth for hdr and denoise"

All HDRx + denoise. These are SHARP. Sharpening is not mentioned and I wonder if the denoise causes the added sharpness without artifacts.

And another by Stan, that I can't locate now. It was the famous Lady in White. Best yet.


Will this HDR be a standalone executable once complete? I mean without hosting Visual Basic and the IC extras?

Here's the interface I see.


The captured tiffs will not load into Adobe or DXO graphics. They do display with Microsoft Photos and Vegas, as 8bit??

Vegas reports it as a BMP.

https://drive.google.com/file/d/11nU...ew?usp=sharing

They must be BMPS, just renamed one and it loaded into Photoshop.

[Mattias Norberg]

Hello

Big shout out to Stan, Bruce and Mattias along with anyone contributing the HDR effort.

The two memorable images in the recent posts for HDR were these:

post#797 by Mattias. "some examples i did not use davinci resolve for color correction only avisynth for hdr and denoise"

All HDRx + denoise. These are SHARP. Sharpening is not mentioned and I wonder if the denoise causes the added sharpness without artifacts.

Thanks

yes the Avisynth Denoise script does much Sharpening on the film you find the script here at POST Nr: #720 it´s modified VideoFred Avisynth script
Videofreds original avisynth script Denoise too much for my liking i want to see Film grain

if it sharpen to much you can lower exampe this line
USM_sharp_ness=28 USM_radi_us=3
to maybe
USM_sharp_ness=18 USM_radi_us=3

this 3 functions in the Avisynth Denoise script does the Sharpening

unsharpmask()
NonlinUSM()
sharpen()

in the last line in the denoise function i overlay the original clip with the denoised + sharpen clip so i get back little Film grain

if you change the

return b22
to
return b2

then you get the denoised + sharpen clip without the overlay of the original clip = then no Film grain or very little


the Avisynth Denoise script is very slow about 0.3 fps or was it even slower but it digs up details very good

[Stan Jelavic]

Thanks David,
Glad you got it working.

Are you asking for padding the image numbers in the app? Just double checking. Yes, that would be easy.

For the takeup you can change the tension in MSP code:
if(P1IN & BIT5) //speed1 check
{
CCR0 += 84; // 1 seconds per frame
//enable_cap = 0; // cannot run capstan in fast mode
if(counter == 100)
{
P1OUT = P1OUT | 0x02;
}
if(counter == 500) -------------------------> Increase the value for higher tension
{
P1OUT = P1OUT & 0xFD;
}
if(counter == 3800) // Turn cam trigger on
{
P2OUT = P2OUT | 0x80;
}
if(counter == 3810) // Turn cam trigger off
{
P2OUT = P2OUT & 0x7F;
}
}

[Stan Jelavic]

Mattias, the sensor used in the UX226 has high dynamic range
https://www.sony-semicon.co.jp/produ..._CQJ_Flyer.pdf


Actually noticed some reference to the on board HDR in the paper... Remember seeing it somewhere when Bruce asked about it but forgot where I I saw it.
Serendipity I guess

Another issue is the monitor limited dynamic range. Just increasing the camera dynamic range will not do it I assume. What was your experience with the 4.4um? How did you view the images. Do you have some that you can share?

[Mattias Norberg]

Mattias, the sensor used in the UX226 has high dynamic range
https://www.sony-semicon.co.jp/produ..._CQJ_Flyer.pdf


Actually noticed some reference to the on board HDR in the paper... Remember seeing it somewhere when Bruce asked about it but forgot where I I saw it.
Serendipity I guess

Another issue is the monitor limited dynamic range. Just increasing the camera dynamic range will not do it I assume. What was your experience with the 4.4um? How did you view the images. Do you have some that you can share?

Ok

here is a google drive link to 3 .tiff that i did use my DFK 23G274 Camera to capture it has Pixel size H: 4.4 µm, V: 4.4 µm
one of the files is HDR_Enfuse.tiff that i did use Enfuse 4.2 and input was Low.tiff High.tiff
https://drive.google.com/drive/folde...Z3?usp=sharing

Low.tiff High.tiff i did use Fast Gpu Debayer to debayer them original is in Y800 format

i remember when i did get my DFK 23G274 Camera i was disappointed about the dynamic range

i did have some super 8mm film that did have very dark film scenes with candle lights it was then i did start to investigate about hdr

one more thing if i remember right Fast Gpu Debayer did better on having the Highlight in check so it does not bleed out more that's why i use Fast Gpu Debayer and not IC capture´s own debayer

[Stan Jelavic]

Thank you Mattias. I analysed the HDR script a bit using your pictures. It is on a border of art and technology.
Here is what I came up with so far.
Read this first:
http://avisynth.nl/index.php/Overlay

Use HDRAGC to increase the gain of high exposure to emphasize the shadows.

The bright areas get overexposed. But they are overexposed anyways.
Multiply with the lo image combined with the high mask.
This darkens a bit the shadows but adds the details from the lo image to
overexposed areas. See the window area.


Use difference with a mask. Where the diff is high (windows) no change, but enhance
dark areas (low difference). The mask controls that.


Combine the multiply and difference. Dark areas are taken from difference and
light areas from multiply. Inverted multiply mask can control which source to pick.


The window area derived from multiplication is still overexposed.
Use lo image which has well defined window to add more detail to that area.
To select the window area only use the combined image softlight blend.
The softlight will select the window area because with softlight only areas that are
above 128 value are brightened. Shadows do not have much impact since they are way below 128.

Now blend the softlight image back with the combined image
.


At this point whoever did this decided that the shadows are too bright used the Darken
overlay to tone them down.


This is just HDR1. Will check HDR2 also. With this image the procedure more or less works.

But it fails with some of my clips. Now that I understand the script a bit better may help me figure out why.