General > General Technical Chat
Can ordinary humans enhance blurry pictures yet ?
antenna:
The only way to add information and reduce noise in a photo is it have multiple frames from the same perspective (of an unmoved object, like a guy on a security camera that stood still for 12 frames) and average them such that the random noise cancels and the similar information reinforces itself. Image enhancements other than adding information also exist, but those are based of manipulating the existing data, for example, using mathematical convolution masks.
Someone:
--- Quote from: SilverSolder on March 08, 2021, 03:10:32 pm ---We are not yet seeing sound being picked out of the image... I'm looking forward to the live demonstration! :D
--- End quote ---
It has been demonstrated, there is nothing preventing these methods being done live (other than funding/development effort):
--- Quote from: antenna on March 08, 2021, 04:54:29 pm ---The only way to add information and reduce noise in a photo is it have multiple frames from the same perspective (of an unmoved object, like a guy on a security camera that stood still for 12 frames) and average them such that the random noise cancels and the similar information reinforces itself.
--- End quote ---
It certainly does not require multiple images of a still/unmoved object. Reconstruction from multiple perspectives/moving scene is usually superior. Some of the details are in the above video, but "enhance" from moving objects has been the traditional use case.
cdev:
Inverse methods or finite element analysis is what is used. If you search on those keywords you'll find a lot of stuff.
I'm pretty sure.
cdev:
It depends on what one wants to enhance. one needs a modern GPU with Cuda for many of the AI video enhancement processing tools.
CatalinaWOW:
The original posters problem, improving resolution of a single static image has some limited possibilities. As mentioned earlier various contrast enhancement techniques can improve the interpretability of the image. Since best results are somewhat subjective best results are usually a trial and error process rather than algorithmic. All of the big image processing programs have ways to manipulate the brightness histogram, usually in a variety of non-linear ways and also analogous to old darkroom dodging and burning in processes these changes can be varied over the image.
If the characteristics of the camera system that generated the image are known there are possibilities in deconvolving the optical transfer function of the camera. Magic doesn't result but you can get some very noticeable improvements. If you still have the original camera the OTF can be estimated using a family of test images. Note that to get it right the camera has to be set up exactly the same way it was in the original picture (f/no, zoom and everything.).
A bit of googling can give you a fair understanding of the process. Actual implementation is always a bit messy.
The hollywood representations are about equally realistic as the matter transmitters on the Starship Enterprise.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version