How to recognize text in an image?

Question

Pattern recognition

You need to write two programs.
The first program processes images arriving at the server and extracts from them those images that have inscriptions.
The second program processes the images arriving at the server and selects from them those images on which there is a print.
I have rather meager knowledge in this subject, so I don’t know where to start and in what direction to move. The level of programming in low and high level languages ​​for PC is high.
Please tell me where to start and in what direction to move then?

Answer 1

Recognizing objects in an image is a tricky topic. It is necessary not so much to be able to program, but to have knowledge in different areas of mathematics.
On Habré there is a good series of lectures from Yandex and one of them is devoted to your topic:
Image and video analysis. Detection of text on ...
Although sometimes it is easier to detect the presence of an object than the morphology of objects on it.
If the author clarifies the details, then a more detailed explanation can be given.
Since text recognition (OCR) is not required, this is probably closer to the topic Image Search by Content (CBIR)
Clustering images without taking into account the objects on them should be easier than the OCR task.
Read this document in read the material in general from the site courses.graphicon.ru .
In general, your task boils down to the following:

1. collect a minimum sample of all kinds of images (for starters, 50 pieces)
   
2. manually classify what to weed out and what to keep
   
3. the algorithm must be able to extract a vector (set) of some metrics from images
   
4. enter the useful component (necessary for classification) into the "dictionary".
   
5. compare each image with the "dictionary" and make a decision about screening
   
6. run the algorithm on a small image base, checking for weaknesses
   
7. improve the algorithm and test again on a larger sample