Copy link to clipboard
Copied
Hello,
[Excuse me for the mistakes, as a non-native speaker]
I would like to know if a picture is real (even with filters) or entirely drawn. Nowadays with modern tools we can easily draw a complex picture starting from pixel number one and use sophisticated algorithms to continue the drawing. And in a near future it might be beyond our natural skills to know whether or not a picture is real (even a video or a game),
I've read about the error level analysis (ELA) which is looking into the different levels of compression artifacts. It has shown some interesting results but is also said to be a non objective tool for this use and has some limits (impossible to use with PNG...),
I've also read about the pixel counting, thus to look for abnormal, unnatural, breaks in the range of colors used,
Also about filtering a picture black and white, setting high contrast and counting on the one hand the white pixels and on the other the grey/black ones to get a ratio,
Is there relevant methods regulary used in the first steps of this kind of analysis,
Thank you
Copy link to clipboard
Copied
What kind of images do you have in mind and what kind of editing are you trying to isolate via image forensics.
I would like to know if a picture is real (even with filters) or entirely drawn.
A rendering of a 3D model is still a »picture«, just like a painting may be a »picture«, what exactly you mean by »real« may need clarification; do you mean to determine if an image is a photographically recorded image as opposed to a photorealistic image?
I've read about the error level analysis (ELA) which is looking into the different levels of compression artifacts.
Compression can be avoided in many workflows (at least for »non-moving« images) so it barely seems the ultimate determiner.