What materials or information can be studied to solve this problem?

Question

Machine learning

spoiler

Answer 1

There is something I cannot understand. "I don't understand the topic of machine learning and the like. I've never worked with this topic" And then where could the task that you NEED (??) solve come from? I also don’t understand absolutely chemistry, but I understand that in order to solve problems, you need to take a textbook and teach-teach-teach ... and only then take on tasks. Moreover, if the task is not entirely trivial-school, then it’s pointless to study something “just for it”, you need to develop some common base, and only then dig in the right direction. Or do you think that machine learning is so trivial that everything is different here? And is it enough just to "look" at something? Very "self-confident" (to put it mildly).
And on the merits of the question - well, start here:
https://ru.wikipedia.org/wiki/Machine_learning
is there and there is a list of references.
or from here
https://ru.stackoverflow.com/questions/Books-and-learn...
or from here:
https://ai-news.ru/2018/10/podborka_resursov_po_ma...
And then follow the links. After about six months or a year of hard work, you can start solving the proposed problem.

Answer 2

The problem contains links to educational materials.

Answer 3

If you do not use AI, then post such tasks in half a year. Start learning programs and algorithms.
Not sure what Docker is, but I would suggest an elementary solution. There are 3 colors: background, cells, and text. So you can do without AI.
Please provide at least one picture.

Answer 4

I agree with the answer that you don't need AI. you're not asked to recognize text, you need to determine if the pixel is for background or text. if the test was represented by the exact value of the color, then everything is elementary. practically the border of text and background will be blurred. then you will be able to determine with a certain probability whether the pixel belongs to the text. digital filtering, in the simplest case, a filter, is quite suitable for this. it can be teachable, those with tuning parameters. Walsh and Haar decomposition, wavelets, are also suitable. if neural networks, then try convolutional networks. in general the task is not set correctly.