Critical performance decrease while capturing from sepia filtered backgrounds

[hamad12a]

I have experienced a significant decrease in performance while using it to capture LaTeX formulas from the pdfjs reader, on which I have applied sepia (0.1) and brightness (0.9) filters for frequent reading.
I didn't notice the issue even with increasing temperature and using GPU (at least remotely) until I tried it on a normal PDF reader without background filters (black and white).
Mathpix snipping tool doesn't have this issue. What could be enhanced or added so that LaTeX-OCR behaves naturally?