Field guide

How a text detector finds words in pictures

This site finds lettering with arithmetic on the pixels — contrast, stroke width, baseline — rather than a trained model, and asks you to confirm every box. This page is about how trained detectors work, because knowing what they look for is what makes a box you check or draw yourself a good one. Finding a word and reading it are separate jobs, and a region can look like lettering even when a model cannot reliably name the characters inside it.

Detectors search for aligned strokes, contrast and word-like geometry across multiple scales. Fences, windows and patterned fabric can produce the same cues.

Recognition confidence helps reject false regions but can also miss stylised, curved or low-contrast lettering. Inspect any mask against the pixels rather than trusting a threshold alone.

Removal requires a mask wider than the visible dark stroke because outlines, shadows and antialiasing occupy neighbouring pixels. Too wide a mask consumes genuine picture detail.

Keep the mask and the fill reversible until review. A plausible patch can hide the fact that the mask crossed a face, product edge or legally important label.