[Yuikoro]

Just looking for suggestions if there are any out there.

I've gotten pretty far with processing my image to get rid of all the unnecessary bits, and extracting just the letter I'm looking for. However, accuracy is terrible. I'm 99% sure the font used is Open Sans, so I've trained tesseract with that font.

What else can I do to improve accuracy?

Here are my images:

Starts off as this


Becomes simplified to this


Is finalized to this


Any ideas on what I might be able to do?

Note: The 'h' is in different areas of the image because the 'h' key moves, but it does not affect the lighting of the image, but may affect image processing at a pixel level due to screenshots working at 96 DPI in a 1920x1080 background. Within my code, I crop to just the important bits (where the 'h' key is bound) making it 234x249, and I upscale the resolution to 300DPI.

Thanks in advance

[darkscrpion]

Just dont play with tesseract ocr in C#, work with it native in C++. Tesseract is not too strong alone and fail too much. Search for the project openalpr in github (sorry cant post links yet) and study the source code from this project. Maybe give you some ideas...