1. Reminder: Please use our affiliate links for holiday shopping!

"Superresolution" with the A7 II

Discussion in 'Video to Share' started by VLReviews, Mar 17, 2015.

  1. VLReviews

    VLReviews TalkEmount Regular

    33
    Mar 16, 2015
    Germany
    Benjamin
    I found this interesting video two weeks ago:



    What the author does is taking ~20 images hand-held and stacking them, creating what he calls a "superresolution" image. The process is explained in details in a blog post here:

    http://petapixel.com/2015/02/21/a-practical-guide-to-creating-superresolution-photos-with-photoshop/

    What is clear is, that you can cancel out noise on stationary subjects with this method. And it's pretty cool you can do it hand-held with the A7 II, mainly thanks to the IBIS. I am not wuite sure that the produced images actually show a higher resolution, though. At least the ones he shows. They are mostly free of noise and slightly sharpened, leading to higher micro-contrast.
    I think it is physically possible to achieve higher res if the stacking in PS works absolutely perfect and each stacked image is 100% sharp and slightly misaligned by fractions of a pixel in relation to the other images in the stack - 0.5 pixels would be good, or 9.2, or 50.7. Full pixels don't really matter, as long as there's a number behind the dot. Then, scaling up and aligning them should actually give you sub-pixel detail. Pretty cool idea!


    What do you think?
     
    • Like Like x 1
  2. xXx1

    xXx1 TalkEmount All-Pro

    Jan 15, 2013
    Yes, noise is statistical thing. Oversampling reduces noise pretty efficiently as it tends to cancel itself when multiple samples are taken.

    For resolution the situation is much more complex. There is the sensor resolution, lens resolution, camera shake and subject movement. I think that noise reduction without usual tradeoffs is responsible for better images. Anyway do we need more than 24 MP resolution?

    Bad poit is of course need for multiple shots. Imagine 1/60 s becoming 1/3 s exposure time.
     
  3. VLReviews

    VLReviews TalkEmount Regular

    33
    Mar 16, 2015
    Germany
    Benjamin
    In everyday life: Certainly not. My 5T's 16 MP is more than enough for me. The 10 MP my former cam brought were, too. But for some situations it might still be fun to have. Take - for example - a night shot of the city without a tripod (and without the A7s ;) ).
     
  4. ztryfe

    ztryfe TalkEmount Veteran

    224
    Aug 19, 2014
    Mexico
    Vic
    This seems quite similar to stacking on astrophotography with dark frames, you take a long exposure, then put the cap back, take another exposure, then blend removing the dark frame, taking out only the noise from the initial exposure, increasing detail in many cases (of course, its much more complicated than that, this is just a general description)
     
  5. WestOkid

    WestOkid TalkEmount All-Pro

    Jan 25, 2014
    New Jersey, USA
    Gary
    Dark frame subtraction does not increase the resolution. In fact it only works for long exposures where fixed patterns of noise on the sensor occur. This is primarily due to hot pixels, stuck pixels, bright pixels and dead pixels. These are particularly problematic in astrophotography because they can look like stars or other light sources in the sky. That type of noise has nothing to do with the actual photograph and thus can be identified by dark frame. The noise being removed here is luminance and color noise which is a direct byproduct of the bright, dark and colored parts of the scene. In this case the noise is being replace with information not removed.
     
    • Like Like x 1
  6. ztryfe

    ztryfe TalkEmount Veteran

    224
    Aug 19, 2014
    Mexico
    Vic
    I should have elaborated more, for me it rang a bell with the dark frame subtraction, but applied to color and addition. Its a very interesting technique. Thanks for the additional explanation and guidance Gary.
     
  7. Lisandra

    Lisandra TalkEmount Veteran

    216
    Jan 28, 2015
    So...I tried this bad boy technique today and, its not as easy as it says there. First of all and most important, You have to have the latest and greatest PC to even think about attempting this. I have an extreme i7 clocked at 4.5 ghz with 32 gb of ram and ram usage at one point was at 93%. Scratch disk made a file that measured 36GBs!!! A PC with say, 12 gb of ram would take a good 5 minutes to stack them, another good 10 minutes to do the 200% resize and over half an hour just to align them. Converting it to a smart object might be an hour long affair.. If you have an i5 with 8gb of ram just stop reading, this is not for you...

    which brings me to the next point, photoshop wont magically align them perfect like the video. thatll maybe happen 3 out of 10 times. So the thing is you have to move just enough between shots so that subpixel averaging actually works but not so much that photoshot cant figure out how to align them. You can manually align them of course but you know... Just know that if theyre not aligned PERFECT, this just wont work. So after all of that if it works then yaaaay, it really does work! but more often than not, a small misalignment or not enough subpixel info will not give you a better result than just taking a single image and doing a 200% resize and sharpening...
     
    • Like Like x 1
    • Informative Informative x 1
  8. WestOkid

    WestOkid TalkEmount All-Pro

    Jan 25, 2014
    New Jersey, USA
    Gary
    Lisandra - Based on your CPU and Ram config, I would imagine you have an SSD and a decent video card? Photoshop uses GPU when it can(btw -next LR will too). If you have those things then I won't even bother. That will not be a fun experiment. I have the 16gb, a GTX 650 ti, and a SSD, but I am running an older 6 core AMD.
     
  9. VLReviews

    VLReviews TalkEmount Regular

    33
    Mar 16, 2015
    Germany
    Benjamin
    I tried it with my NEX and can second that the alignment in Photoshop doesn't work precise enough on most images. You have to adjust them by hand. Also, every single image really has to be tack sharp, which usually isn't the case even with OSS/IBIS. This caused me to delete about 50% of the images in the series. Stacking was slow, in the end, but even though I scaled it up 3x instead of 2x, it worked quite well on my i5 with 8 GB of ram :D Maybe the SSD helps to speed up the swapfile. Anyway, I can confirm that using a smart object is not a good idea! Waaaayyyyy too slow.
     
  10. Lisandra

    Lisandra TalkEmount Veteran

    216
    Jan 28, 2015
    2 TB ssd yes, and a gtx 8 series. Photoshop is more ram intensive than anything else, so having ram cards clocked at say 1866 mhz will speed up things much more than getting a super multicore setup. Ive seen i7s clocked at 4.8 ghz with 1333 mhz sticks that are dog slow on PS. You can get get away with it if youre rockin an ssd, but its not gonna be like the video fast. heck mine wasnt fast like the video, so either he skipped ahead or he has an unknown futuristic pc configuration.
     
  11. lenshoarder

    lenshoarder TalkEmount Veteran

    297
    Feb 7, 2012
    This isn't anything new. It's been done in science for decades. It's even builtin to the NEX. It's the various multishot modes like handheld twilight. Why aren't people just using those?
     
    • Like Like x 1
  12. Lisandra

    Lisandra TalkEmount Veteran

    216
    Jan 28, 2015
    because handheld twilight does the average thing for noise only and not resolution. And jpeg only too
     
  13. lenshoarder

    lenshoarder TalkEmount Veteran

    297
    Feb 7, 2012
    You can't do one without the other. What's being done is simple image stacking. It's not operating just on the noise. It's happening across the image as a whole. What's happening is that the noise is being averaged out. By doing so the image becomes clearer. The signal to noise ratio is being increased. That's the whole point of noise averaging. This is exactly what handheld twilight mode does. HHT works very well with moving objects as opposed to doing it in PS.
     
  14. Lisandra

    Lisandra TalkEmount Veteran

    216
    Jan 28, 2015
    Ok but sonys purpose with hht was noise wise. The hht files arent any more detailed than regular files. What i mean is that it's not like oly or the video where thy meant to get higher resolution