Author Topic: Can ordinary humans enhance blurry pictures yet ?  (Read 4781 times)

0 Members and 1 Guest are viewing this topic.

Online MathWizardTopic starter

  • Super Contributor
  • ***
  • Posts: 1772
  • Country: ca
Can ordinary humans enhance blurry pictures yet ?
« on: March 07, 2021, 08:40:06 pm »
I'm working on a repair , and I lost some of the notes I took when I was making a schematic. But I found some old picture of the PCB on the web. And I found a missing resistor not in my notes, but from the picture, I can just make out the resistor is actually just a jumper.

But that reminds me, since the 90's as a kid, in all kinds of TV and movies, people are always making blurry pictures look better.

Are we there yet ? I don't see any button in windows 10 picture viewer for it, or is some math for it actually built in?. Or can I get a freeware program for this yet ???
 

Offline MikeK

  • Super Contributor
  • ***
  • Posts: 1316
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #1 on: March 07, 2021, 09:34:32 pm »
No, and we never will be.  There is no information to extract from a single pixel.

Alternatively, just say "Enhance!"
 

Offline Gyro

  • Super Contributor
  • ***
  • Posts: 10172
  • Country: gb
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #2 on: March 07, 2021, 09:35:03 pm »
Ah, the old Hollywood zoom into a tiny fragment of an image and enhance.  ;D

You might try importing the image into something like Gimp and trying a little sharpening and contrast enhancement, but you're not going to be able to reveal detail that isn't there, just pep it up a bit.
Best Regards, Chris
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #3 on: March 07, 2021, 10:04:59 pm »


There are plenty of ways to enhance resolution -- but doing so is increasingly questionable, statistically speaking.  We can use an L1-norm interpolation for example, which gives lower noise results from surprisingly sparse inputs; we can also use neural networks to fill in what's likely missing, but this is contingent on the data the neural network was trained on -- that is, if it was trained on pictures of buildings, it's more likely to find stucco or brick or wood textures (or some hallucinogenic blend of everything!) when zoomed in, regardless of what the actual texture was (e.g., maybe it's actually a smooth metal surface, but the optical noise of the camera is being interpreted as real texture).

The worst outcome is trying to speculate on something that well and truly has no information to support it.  We could take a ten pixel image and expand it full screen.  Obviously, the number of textures, and objects, that can be contained in that full size image, vastly exceeds that of the source image -- and just about any of them are as statistically likely as another, given that the overall colors are approximately right to match the original image when this image is reduced.  Today, we can very easily create images that are a shockingly similar to a speculative organic process, as if using ones' imagination to fill in the blanks.  It's like a dubious witness, testifying of a story, which they seemingly created from whole cloth; whereas the scene they were actually witness to, was perhaps a few seconds of shouting.  And similarly, if we are to put too much trust in such a representation, we may find we are creating some very bad outcomes indeed (e.g. wrongfully imprisoning someone on false evidence).

Okay, so going from basic image processing, to legal proceedings, is a rather abstract step, I admit; but to the extent that photographic evidence might be used in such proceedings, it's very applicable.  And more broadly, we can guess that analogous processes occur within a witness's mind -- that is what I'm getting at.

So -- if your image isn't very blurry, maybe an extra 20 to 100% of resolution could be drawn from it, maybe even 200 or 400% in extreme cases; but fine details will very quickly become questionable at that rate, especially if the algorithm is prone to synthesizing meaningful details (like text, or faces) where there isn't necessarily anything of the sort to begin with.

As for actually trying it to see, I'm not sure offhand where you can find such a processor?  There may be some online tools, worth looking around to see.  I'm sure Photoshop has all manner of clever filters available.  Or you could download some of the tools these are based on, which may require some degree of licensing, or building (e.g. from source code, or the tools just aren't very convenient e.g. CLI based), or cooking up utterly from scratch (you can do all manner of machine learning yourself, for free; but, it is an entire field to learn about, good luck -- and truly, good luck, it will surely prove very useful/profitable to put the time into, if you choose to!).

Tim
« Last Edit: March 07, 2021, 10:08:35 pm by T3sl4co1l »
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: cdev, Gyro

Online coppercone2

  • Super Contributor
  • ***
  • Posts: 11336
  • Country: us
  • $
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #4 on: March 07, 2021, 10:09:52 pm »


replaced with something a little more PC
 

Offline DrG

  • Super Contributor
  • ***
  • !
  • Posts: 1199
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #5 on: March 07, 2021, 10:33:42 pm »
It is an interesting question. I rather agree with a lot of what has been said, which I would paraphrase as "you can't get something from nothing, unless you make something up".

But, there are some interesting approaches. As it turns out I was just reading about these youngins and some work they are doing http://pulse.cs.duke.edu/

It is an interesting approach to go "in the other direction" see the brief video.

So, take for example a blury frame of a video from a security camera that shows a car's licence plate - not enough detail to read. Conventional enhancement is probably not going to get you where you want to go. But, what about if you construct the end product (the blurry license plate) starting from many high resolution frames of many license plates. From there, you could start to evaluate the probabilities of different alphanumeric characters, after the lowering of resolution, toward getting some useful information. Neat stuff.
« Last Edit: March 07, 2021, 10:42:50 pm by DrG »
- Invest in science - it pays big dividends. -
 

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #6 on: March 07, 2021, 11:03:02 pm »
Okay, so going from basic image processing, to legal proceedings, is a rather abstract step, I admit; but to the extent that photographic evidence might be used in such proceedings, it's very applicable.  And more broadly, we can guess that analogous processes occur within a witness's mind -- that is what I'm getting at.

So -- if your image isn't very blurry, maybe an extra 20 to 100% of resolution could be drawn from it, maybe even 200 or 400% in extreme cases; but fine details will very quickly become questionable at that rate, especially if the algorithm is prone to synthesizing meaningful details (like text, or faces) where there isn't necessarily anything of the sort to begin with.
True for general cases with a single source frame, but most of the impressive "forensics" enhance effects are from video source with many many frames (and possibly strong aliasing) to support the reconstruction..
But that reminds me, since the 90's as a kid, in all kinds of TV and movies, people are always making blurry pictures look better.

Are we there yet ? I don't see any button in windows 10 picture viewer for it, or is some math for it actually built in?. Or can I get a freeware program for this yet ???
All this tech is getting buried into the cameras, doing the enhance at capture time before saving the already enhanced image for your convenience (or annoyance depending on perspective). Image sensors have plateaued in performance for a while now and the image quality improvements year on year are more driven by image processing than hardware/optics/etc.
 

Offline james_s

  • Super Contributor
  • ***
  • Posts: 21611
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #7 on: March 07, 2021, 11:12:04 pm »
Unfortunately Hollywood image enhancement is like most Hollywood things, it's a special effect, pure fabrication and fantasy and in real life almost nothing works the way it does in movies.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #8 on: March 07, 2021, 11:17:15 pm »
True for general cases with a single source frame, but most of the impressive "forensics" enhance effects are from video source with many many frames (and possibly strong aliasing) to support the reconstruction..

Yes!  If you have additional information, make use of it! :)

There's even a method to recover live audio from even rather poor quality video (e.g. < 60 FPS, low resolution webcam) -- the trick is using very small signals (down in the LSBs of pixels), over large areas, and correlating the changes across that area.  The pixel count recovers your SNR, and the fact that audio waves propagate through space, over small objects in the scene -- plants, paper, wrappers, that sort of thing -- you can essentially map out oscillograms in video.

As far as I'm aware... the technique isn't a general purpose, drop in and go, method -- it requires a lot of massaging to recover useful audio, and obviously is much better with higher framerate, resolution and color depth.  You can appreciate, it might not be obvious how sound waves move across the objects in a visual frame.  And with so little information to start with (noisy LSBs), it's not easy to solve for a correct pattern of wave propagation in the scene (essentially, I would suppose, a plot of delay and gain per pixel), let alone general angles since there might be multiple sound directions besides.  But is it possible, absolutely! :o


Quote
All this tech is getting buried into the cameras, doing the enhance at capture time before saving the already enhanced image for your convenience (or annoyance depending on perspective). Image sensors have plateaued in performance for a while now and the image quality improvements year on year are more driven by image processing than hardware/optics/etc.

The amount of CPU power that goes into a current model cell phone is incredible; they're putting neural network stuff into them, too -- not just GPU accelerated but direct NN/ML hardware too.  Realtime high def augmented reality is well and truly here.

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 
The following users thanked this post: Someone

Offline RJSV

  • Super Contributor
  • ***
  • Posts: 2756
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #9 on: March 08, 2021, 12:14:00 am »
Uh, I'm having (friendly) doubts, first thought:

    What would the wavelength be, of an audio pressure wave, going across that room, on camera? It just seems way way small, and has to be in terms of displacing small surfaces, like a plant leaf.
I need to check that, certainly the displaced surfaces are going to move, like zilch...
But wavelength, at 1khz, I think is roughly around one foot (??).
 

Offline RJSV

  • Super Contributor
  • ***
  • Posts: 2756
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #10 on: March 08, 2021, 12:24:51 am »
Ok, so I looked up the wavelength, it's 1.1 foot, at sea level, 1Khz sound wave. The sound extraction algorithm can look one place, for peaks, contrasted with another location, within some 1.1 foot space, to extract data in a phased manner, (but assuming you have the propagation direction to try).

  It's WILD, I doubt, but if I had to bet my life, it's probably something  that can be, perhaps low frequencies only (below 100 Hz).
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #11 on: March 08, 2021, 12:35:53 am »

The sample rate cannot be higher than the frame rate?  so 30Hz out of a web cam?   by Nyquist, we should be able to detect 15Hz!  :D
 

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #12 on: March 08, 2021, 01:09:53 am »
There's even a method to recover live audio from even rather poor quality video (e.g. < 60 FPS, low resolution webcam) -- the trick is using very small signals (down in the LSBs of pixels), over large areas, and correlating the changes across that area.
There is a wealth of information hiding inside most images, but it comes down to very specific characteristics of the image sensor as you noted.
As far as I'm aware... the technique isn't a general purpose, drop in and go, method -- it requires a lot of massaging to recover useful audio

But then, lol, the uninformed come out to announce their superior fundamentally incorrect understandings again.
The sample rate cannot be higher than the frame rate?  so 30Hz out of a web cam?   by Nyquist, we should be able to detect 15Hz!  :D
Or you could make an equally misleading performance claim about a camera at 60Hz frame (field) rate, that it has 8GS/s
https://www.photonics.com/Articles/133-MP_CMOS_Image_Sensor_Eliminates_Bulky_Optics/a57239
OMG 4GHz nyquist!!!!@
No, not really as the sampling is not evenly spaced in time.
 

Offline RJSV

  • Super Contributor
  • ***
  • Posts: 2756
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #13 on: March 08, 2021, 02:01:14 am »
Ok,pls excuse, return to Enhance of images:
  I am aware of many examples, of course there is phantom limb, or false sense (of lost limb).
  For your question, human brains are ready to 'fill in' for a lost or defective sense. So the waving of the hand, across in front of the face, can cause the brain to 'fill-in', again for the missing or flakey ACTUAL view.
  This optical, and neurological process, I believe is termed (informally, as I am not a PRO), termed as
'Visual phantom limb'.
That's where the person has a vague impression or sight, blurred blob, only approximately where the actual real hand is located.
  So, in this case you maybe can't trust the image, but there possibly are these internal mechanisms that have more validity.
In my example above, there is NO visual connection, rather all the 'input' being the motor neurons and muscle info, being used in this strange process.
  If you think about it, that arm of yours has moved across your field of view, many thousands or millions of times. So in absence of vision, that brain tries to supply or synthesize that view.
It's a huge field, but it's interesting to study (neurology) in the context of the modern processors and image processing math.
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #14 on: March 08, 2021, 02:23:33 am »
There's even a method to recover live audio from even rather poor quality video (e.g. < 60 FPS, low resolution webcam) -- the trick is using very small signals (down in the LSBs of pixels), over large areas, and correlating the changes across that area.
There is a wealth of information hiding inside most images, but it comes down to very specific characteristics of the image sensor as you noted.
As far as I'm aware... the technique isn't a general purpose, drop in and go, method -- it requires a lot of massaging to recover useful audio

But then, lol, the uninformed come out to announce their superior fundamentally incorrect understandings again.
The sample rate cannot be higher than the frame rate?  so 30Hz out of a web cam?   by Nyquist, we should be able to detect 15Hz!  :D
Or you could make an equally misleading performance claim about a camera at 60Hz frame (field) rate, that it has 8GS/s
https://www.photonics.com/Articles/133-MP_CMOS_Image_Sensor_Eliminates_Bulky_Optics/a57239
OMG 4GHz nyquist!!!!@
No, not really as the sampling is not evenly spaced in time.

Let's just say that detecting audio by looking at images is probably in the "challenging" territory!  :D
 

Offline Tomorokoshi

  • Super Contributor
  • ***
  • Posts: 1212
  • Country: us
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #15 on: March 08, 2021, 04:12:17 am »
 
The following users thanked this post: T3sl4co1l

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #16 on: March 08, 2021, 04:49:54 am »
But then, lol, the uninformed come out to announce their superior fundamentally incorrect understandings again.

You could really stand to leave people a benefit of a doubt...

I read it as sarcasm--well, I hope I read it correctly as sarcasm, at least. :P

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline Someone

  • Super Contributor
  • ***
  • Posts: 5155
  • Country: au
    • send complaints here
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #17 on: March 08, 2021, 06:19:25 am »
But then, lol, the uninformed come out to announce their superior fundamentally incorrect understandings again.
You could really stand to leave people a benefit of a doubt...

I read it as sarcasm--well, I hope I read it correctly as sarcasm, at least. :P
If people choose to post nonsense in a public forum and think they're clever, expect to be corrected. Its about as silly as the often repeated "but scope screens only refresh 60 times per second", a measure entirely unrelated to the sampling rate of the signal acquisition, or other characteristics also measured in Hz.

With a good model of the sensor we see all sorts of clever things like stabilization, rolling shutter motion compensation, flicker removal, etc. All built into the live processing path on photo and video cameras. Mostly proprietary stuff and completely invisible to the end user.
 

Offline T3sl4co1l

  • Super Contributor
  • ***
  • Posts: 22436
  • Country: us
  • Expert, Analog Electronics, PCB Layout, EMC
    • Seven Transistor Labs
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #18 on: March 08, 2021, 06:58:07 am »
Well, fortunately, pi is exactly three. :)

Tim
Seven Transistor Labs, LLC
Electronic design, from concept to prototype.
Bringing a project to life?  Send me a message!
 

Offline Syntax Error

  • Frequent Contributor
  • **
  • Posts: 584
  • Country: gb
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #19 on: March 08, 2021, 10:39:06 am »
OP: No and yes. There are a few tricks you can try at home such as unsharp mask, edge detection, hysteris stretching, pixel averaging and gausian blur. All of those high tech 1980s digital image 'convolve matrix' tricks have been done in photoshop since the 1990s.
https://en.m.wikipedia.org/wiki/Kernel_(image_processing)

Plus, there are the image stacking techniques used by astrophotographers.
https://www.skyatnightmagazine.com/astrophotography/astrophoto-tips/a-guide-to-astrophotography-stacking/
 
The following users thanked this post: MathWizard

Offline TerraHertz

  • Super Contributor
  • ***
  • Posts: 3958
  • Country: au
  • Why shouldn't we question everything?
    • It's not really a Blog
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #20 on: March 08, 2021, 10:44:34 am »
In youtube, search for "unblur image in photoshop"

You'll be surprised.

"windows 10 picture viewer" ??? Surely you jest?
Collecting old scopes, logic analyzers, and unfinished projects. http://everist.org
 

Offline jonovid

  • Super Contributor
  • ***
  • Posts: 1546
  • Country: au
    • JONOVID
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #21 on: March 08, 2021, 01:25:56 pm »
There's more than one way to skin a cat . my apologies to all cats. but
electronic circuit you are seeking may have other ways to extract what you are seeking.
if its only a missing resistor value, by deduction.  looking at other known circuits of the some or  similar
« Last Edit: March 08, 2021, 01:28:02 pm by jonovid »
Hobbyist with a basic knowledge of electronics
 

Offline SilverSolder

  • Super Contributor
  • ***
  • Posts: 6126
  • Country: 00
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #22 on: March 08, 2021, 03:10:32 pm »
But then, lol, the uninformed come out to announce their superior fundamentally incorrect understandings again.
You could really stand to leave people a benefit of a doubt...

I read it as sarcasm--well, I hope I read it correctly as sarcasm, at least. :P
If people choose to post nonsense in a public forum and think they're clever, expect to be corrected. Its about as silly as the often repeated "but scope screens only refresh 60 times per second", a measure entirely unrelated to the sampling rate of the signal acquisition, or other characteristics also measured in Hz.

With a good model of the sensor we see all sorts of clever things like stabilization, rolling shutter motion compensation, flicker removal, etc. All built into the live processing path on photo and video cameras. Mostly proprietary stuff and completely invisible to the end user.

We are not yet seeing sound being picked out of the image...  I'm looking forward to the live demonstration!   :D
 

Offline Syntax Error

  • Frequent Contributor
  • **
  • Posts: 584
  • Country: gb
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #23 on: March 08, 2021, 03:40:45 pm »
We are not yet seeing sound being picked out of the image...  I'm looking forward to the live demonstration!   :D
I seem to remember --- vary vaguely --- an hypothesis that ancient sounds can be preserved in tacky surfaces in the same way that sounds are preserved in shellac cyclinders and records. Okay, I'm not sure how you would make a machine to extract those ancient sounds, ( petra-hertz? ), but in theory you might just hear the last words of the people of Pompeii.
 

Offline grumpydoc

  • Super Contributor
  • ***
  • Posts: 2931
  • Country: gb
Re: Can ordinary humans enhance blurry pictures yet ?
« Reply #24 on: March 08, 2021, 04:09:16 pm »
In youtube, search for "unblur image in photoshop"

You'll be surprised.
Unbluring is slightly different from enlarging - in the former case the information should be in the picture, just in the wrong place and depending on how the blur is caused there are mathematically sound algorithms for recovering the correct image.

Increasing the pixel resolution of an image needs new information to be extrapolated from what is there. In some (limited) cases (say a line drawing) this can be done surprisingly well but the general case is harder.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf