Some people say that the invention of mosaic is a major "reversing" in the history of human civilization.
It is also said that mosaic is a stumbling block that hinders human progress.
In terms of function, mosaic is a means to cover up the original picture, which has advantages and disadvantages.
Technically speaking, the principle of mosaic is to smear the color of a certain pixel to the entire range and cause the original picture information to be lost.
Therefore, this process is irreversible.
In the Internet age where personal privacy is extremely easy to leak, everyone may use mosaics to cover up sensitive information in pictures.
For example, the ID number, name, and address can be effectively protected for the purpose of information security after being masked with mosaics.
But in the same way, in film and television comics, whenever there are scenes that are inappropriate for children or bloody violence, they are most likely to be accompanied by heavy mosaics.
Even if it is a serious picture, as long as you bring a mosaic, the style of painting will become weird.
Although the old driver's door has long been practiced into the highest state of "Natural HD without code in the heart". But watching a movie with a size, it is always like scratching through a boot, like a knuckle in the throat.
As the saying goes, XX is the first technological productivity. Since the emergence of mosaics in the human world, naturally some people have focused on how to code.
For a long time, software that claims to be able to remove mosaics has emerged one after another, but most of them are malicious virus software that sells dog meat.
However, recently on Github, an artifact that claims to be able to erase the mosaic and reappear the original image has appeared, which has caused heated discussions at home and abroad.
This tool is called Depix, and it has more than 10,000 stars on GitHub within a few days of its launch. So far, the cumulative number of branches has exceeded 1.3k.
What makes it out of the circle is the following rendering:
As shown in the figure, the first line is the text content completely pixelated with a thick mosaic, and no original traces can be seen.
The second line is the content restored by AI. You can see that the content has basically been restored, and the accuracy is very high.
The third line is the original password obtained by a little reasoning based on the second line.
Yes, the text that is the most difficult to be restored, or even theoretically impossible to restore, was restored by Depix after heavy codes!
This means that even if you post sensitive personal information such as your name/ID number/telephone on the Internet in a mosaic, you may be stripped of it by others.
So, how is this incredible AI reduction technology realized?
According to the project author, using Depix to restore the mosaic text to the original image requires only four simple steps:
Take a block of pixels from the screenshot as a selection.
In an editor with the same font settings (including text size, font, color, hsl), paste the De Bruijn sequence with the expected characters.
Make a sequence of screenshots, using the same screenshot tools as possible to create pixelated images.
Execute the following commands:
python depix.py -p[pixelated rectangle image] -s [search sequence image] -o output.png
As we know, coding is a means of image pixelation processing, which is to achieve a blur image effect by degrading the level details of a specific area of the image and disrupting the color blocks.
Linear box filter is one of the more common processing algorithms.
As shown in the figure below, after an emoticon image is coded, it is actually divided into four color blocks, each of which is covered by the average value of the color block, and finally a pixelated expression is formed.
Since the original information is lost, the filter cannot be inverted directly.
Depix's algorithm is to use the linear box filter to process each color block separately, and pixelize each block in the search image to find a direct match.
To put it simply, the technical principle of Depix is to divide the content of the mosaic area into many small blocks, and then use the algorithm to calculate and cover the average value of the content of each small block.
At present, Depix is a tool in a specific field. It is only suitable for pixelated images created using linear box filters, and supports mainly English letters, numbers, and symbols under English half-width.
After testing, most English letters, numbers and symbols can be correctly identified by Depix.
However, during the research process, the developer found that some consecutive characters could not be recognized correctly, so improvements should be made.
Although the output of the Depix algorithm is not perfect, it is quite good compared to other previous technologies.
The news about Depix immediately sparked a heated discussion after it was uploaded on the Internet, and many netizens expressed their deep concern about this technology.
The developer emphasized that the development of this AI project is not to steal information, but to use ECB and plaintext attacks to improve information protection technology.
He also suggested in his blog that for security reasons, please do not use mosaics. The best practice is to directly cover the areas that need to be blocked with black or white stripes, which cannot be restored by algorithms.
Of course, in addition to Depix, there are many open source projects that try to restore mosaic content.
Previously, expansion fans also introduced to you an algorithm called PULSE-it can "draw" a new face based on the blurred mosaic pixels, and the original resolution of the image can be enlarged by 64 times in a few seconds. , Any slag image quality can be changed into high-definition, lifelike images in seconds.
The eyebrows, eyelashes, pores, wrinkles, and hair can all be restored.
As for the popularity of this kind of projects, there is a lot of skepticism on the Internet. These AI technologies are inevitably used by some people for bad or illegal purposes.
It is foreseeable that AI reduction mosaic technology will become more and more mature, and the ethical and legal aspects of demosaicing will become more and more complicated in the future.
GitHub address
https://github.com/beurtschipper/Depix