Locate on-screen text in a screenshot for automation using Tesseract OCR when no accessibility tree is available

Steps

1. Capture a screenshot of the target region, then run `tesseract <screenshot> <outputbase>` for basic recognition (English + page segmentation mode 3 by default)
2. Select the right page segmentation mode with `--psm` when the default general-layout assumption doesn't match your UI (e.g. a single line of text vs. a full-screen layout)
3. Set the recognition language(s) with `-l eng` (or `-l eng+deu` etc.) to match the UI's actual language rather than relying on the English default
4. Choose the OCR engine mode with `--oem 1` (LSTM/neural, generally more accurate) vs `--oem 0` (legacy engine) depending on your Tesseract build's available models
5. Request `hocr` or `tsv` output instead of plain text to get per-word bounding boxes and confidence scores (e.g. `x_wconf`), which is what lets automation code click at a specific word's coordinates rather than just reading the text
6. Treat this as a fallback strategy behind accessibility-API or DOM-based locators — reserve it for legacy/non-instrumented apps where no accessible tree or automation ID exists