Make Printed Photo Captions Searchable Without Changing the Scan

Use the inscription scan as the source

Start with the untouched scan of the album page or photograph back. Make a working crop containing the printed caption and enough margin to avoid clipping letters. Correct the crop's rotation if necessary. Tesseract's documentation identifies skew, borders, noise, and segmentation as factors affecting recognition. Review the text crop separately from the portrait you may want to restore.

A generative restoration can alter a faint letter into another plausible shape. Do not use an AI-restored caption as your only reading source. Keep a clear relationship between the full source, OCR input crop, raw text, and reviewed transcript. Handwriting and mixed scripts may need manual transcription or a different tool; this example concerns printed text.

Run a small recognition test

If Tesseract and the required language data are already installed, the example below reads an English printed-caption crop and creates a text output using a separate basename. Replace the sample filenames. Use the language model appropriate to the actual text; selecting English does not translate another language. ImgHeal does not provide this OCR command or install language data.

• Keep PH-0001-ocr.txt as raw machine output.
• Create PH-0001-transcript.txt for a reviewed reading.
• Record the full source filename in both files.

tesseract "PH-0001-caption-crop.png" "PH-0001-ocr" -l eng

Review names before using them for research

Compare every personal name, place, and number against the unedited scan. Look for letter-number confusion such as O and 0, merged words, and missing accents. Keep doubtful readings marked. For example, a machine reading of 1918 should remain uncertain if the final digit is obscured. Do not let an apparently precise OCR result override what is actually visible.

Keep spelling as written in the reviewed transcript. Put modern spellings and alternate search terms in a separate note so a future reader can distinguish the source from your interpretation. Have another person check the difficult lines. Retain the raw result even when it is wrong, because it documents which proposed reading was corrected and why.

Test search and hand off the evidence together

Search for one verified name using the archive application you intend to use. Some applications index text sidecars automatically; others require importing the text into a description field. Check what yours does instead of assuming that files in the same folder are searchable together. Confirm that a search result takes the reader back to the correct full scan.

Keep the source scan, text files, and image ID together when sharing the archive. ImgHeal can create a separate viewing restoration of the photograph, while the original caption scan supports the research record. Label the restored image clearly and keep the checked transcript accessible beside it. A cleaner portrait should never become the only surviving evidence of a caption.