Hello, We are processing screenshot PNG images from computer display with 96dpi resolution. This is just B/W images rendered with known truetype single font with fixed size, w/o antialiasing or any other subpixel rendering things. Picture structure is clear, opaque and pixelated like listed below, character or glyph always have the same 2D structure:
[image: screen1.png] Main goal is simple, fast and efficient (in time and memory terms) OCR of these screendumps with 100% accuracy. We tried tesseract 4/5 with old Cubic and new LSTM models in default mode with average results (60%), then trained both models with custom font and 288-384dpi upscale, unfortunately final output is still not good - recognition is definitely better around 90%+, but not 100% and additional upscaling increases memory and processing in 10 times. Does anyone know if Tesseract provides some kind of configuration or functionality to do this kind of OCR? Or may be some other open-source OCRs better fit this task, like EasyOCR, OpenCV, OCRopus, GOCR etc.. -- Thanks, Vadim. -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to tesseract-ocr+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/e8e50184-c64d-4808-85e4-7948c93e261bn%40googlegroups.com.
