Hey Guys, I am working on a dll that has high performance and high accuracy OCR imaging (only because it is specifically made for Diablo 3 fonts ). When it's done I am going to release the source code and the dll on here. I have a prototype that works using neural nets to identify different characters.
Fonts planning to identify:
-The font used for item names.
-The font used for most text, (item stats, chat room text, etc.)
-The weapon damage font
-If you want me to add any other fonts to recognize please let me know
Progress:
-Splitting and Refining images: 95% (Only because it's not finalized.
-Neural Network Image Recognition: Recreating system but previous progress showed much potential. 30%
I am planning on releasing it in 32-bit, but anyone can recompile it for 64 bit. It's coded in c++ and its one dependency is opencv, which i will be providing the necessary dlls.
Motivation:
I've seen lot's of profiles that use Tesseract-OCR to do character recognition, and, while it works somewhat, if we at least filter out a lot of the noise it will have a much better chance of recognizing the text. I decided to add the character recognition because I am a computer programmer and I think neural nets are cool so I wanted to try training them to recognize the this text.
Progress Samples:
Original Image:
Filtered( process has since been changed from this older sample):
Filtered Again:
Image Splitting:
Original Image:
Filtered(Latest Algorithm):
Filtered Again:
Image Splitting: