How to denoise an image with text to use it in Tesseract OCR?

Question

Image processing

The crux of the matter is this. There is a binary picture (state number of the car). The image is passed on to the tesseract API for OCR. If there is no noise, then everything works fine - the text is recognized. However, in reality there are always spots in the image. Gaussian blur and morphology operations already applied, small details filtered out. But there are a few spots left that lead to an incorrect result when recognizing characters. Maybe there is an opportunity to somehow configure tesseract? I use all default settings. In theory, it is necessary to select connected areas on the autonomer and filter out those that have a deliberately small area. But how to do it quickly with opencv, I don't know. Thank you.

Add

If someone is interested, then this problem can be solved like this.
Apply the Canny operator to the license plate image, while looking only for outer contours
. Then filter the resulting contours by size, cut them out and recognize each character separately.
There is almost no drop in performance.

Answer 1

I'm trying to do something similar, I need to recognize the numbers on the receipts. There are problems with recognition - small details at the edges interfere. But since I am far from image processing, I don’t understand the phrase “filter the resulting contours by size, cut out”. Is this done with opencv? Can you tell me where I can see how it's done?

Answer 2

"Is this done with opencv? Can you tell me where I can see how it's done? »
Yes, this can be done using opencv
I dashed off an article on IBM DW with code examples if you are interested - knock the link in a personal