Library for working with images in C++?

Question

C++ / C#

I decided to try to write a program for image recognition (for a term paper), or to be more precise, numbers (MNIST database). I read what different people write on this topic and decided that what to choose anyway, how and what - you need to choose a good one, so that later you don’t rewrite the code 10 times and curse yourself with a bad choice.
I haven't decided which IDE to write in either - I have Borland C++ Builder 6 or MS Visual Studio 2008 Pro || 2010 Express.
I would like to hear the answer to the question which library / class to choose based on the following:
1. Simple installation in the described IDE or complex, but well (read: step by step) sorted somewhere on the off site.
2. Speed ​​of work with images
3. Russian documentation is desirable, you can use this option: Russian is a bit outdated (a book, for example, or just a website) and very fresh English
4. Good documentation , not just a couple of lines about each function, but also a description of the arguments, return parameters and
5. Required examples, examples, examples .
PS: if you have seen a comparison of libs for working with images somewhere, please share. I googled only a couple of old articles, and even there there is a lot of subjectivity.

Answer 1

To be honest, I didn’t quite understand, do you need a library for working with graphics, or is there a library that is already sharpened for neural networks?
As for graphics, definitely ImageMagick.
As for your term paper, here is an old article to help you, but very understandable.
Neural networks, image recognition (Magazine Programmer No. 10, 2001)

Answer 2

OpenCV is the first thing you need to pay attention to, there are many good examples in the library itself, it should be enough for almost everything

Answer 3

Well then ImageMagick, fast, reliable and constantly evolving, is widely used on large projects.

Answer 4

I will not say anything about the library, but I will tell one story. My good friend earned money by writing programs for diplomas and term papers. One girl approached him with a similar problem - she was asked to write a text recognition program.
The first thing he did was split the text into lines, analyzing the dot density in each line of pixels.
The second step was to break the lines into characters, and he took into account that in modern fonts the width of different characters can be different.
Finally, he showed the found characters to the user and offered to enter from the keyboard a letter or character corresponding to the image. Thus the program was trained.
The result was amazing- after training, the program showed better results than any text recognition systems existing at that time (1995).
The program had two drawbacks - a rigid binding to the font(s) on which the training was performed. The more fonts the program understood, the lower was the recognition accuracy.
Many years have passed since then. It is a pity that the friend never returned to this topic. His simple program would even now make sense for recognizing large large text printed in one font, such as books.
By the way, I also don’t understand why the author of the question needs a library for working with images - use monochrome BMP at the input of the “recognizer”. The meaning of the course paper is not in working with graphic libraries, but in text recognition.