I'm reaching out for some friendly advice on a project using Tesseract to extract room names from floor plan images. I'm pretty confident my post-processing is on point, but I'm stuck on the tricky parts: pre-processing and model configuration. My Workflow:
Pre-processing: I've tried grayscale conversion, resizing, and even some adaptive thresholding magic, but it seems like I'm missing something for consistent results.
Model Config: I'm using PyTesseract with the "Sparse text" (psm 11) config hoping to catch everything, but maybe that's not the best approach?
Post-processing: Filtering out clutter and matching names with a dictionary feels pretty good.
Result
Out of 50 test images, I get stellar results on 21 images (most room are detected).
But then there are these quirky cases:
Crystal clear plan, but only one lonely room detected. 
Accuracy also takes a dive here, and I suspect the culprit might be the colorful background. 
So, wise OCR gurus, I'm humbly asking for your insights! ♀️
Which pre-processing techniques should I befriend to tame these tricky images? Is there a better Tesseract config hiding out there? I'm open to exploring different approaches! Any general floor plan OCR wisdom would be gold.