Scanned PDF OCR: PDF24 & Sejda Test Outputs

PDF24 and Sejda: what was searchable after OCR?

Both online tools returned a downloaded PDF with extractable text and matches for three test strings. The extracted symbol rows differed from the source. Proofreading was still needed.

One clean, single-page English fixture; one Browser run per tool. No accuracy percentage, overall winner, editable-document result, desktop result or tested exact match is established.

Same source pixels, different input formats

PDF24 received the original PNG. Sejda received an image-only PDF containing the exact same decoded RGB pixels. The PDF input had no extractable text and none of the three search markers before OCR. PNG and PDF wrappers can affect processing and page sizing, so these runs are a small workflow comparison, not a controlled accuracy benchmark.

The synthetic page is 1654 × 2339 pixels at 200 dpi. It contains invented English text, numbers, punctuation, simple text rows and smaller print. No personal or customer document was used.

Download source PNGDownload image-only PDFDownload known source textInspect same-pixel validation

Original symbol row: Symbols: + - = ( ) [ ] % & @ #. These are exact Poppler extraction examples; spacing depends on the extraction method. Neither example establishes an overall error rate.

PDF24 Online OCR: saved output and evidence

This is the Online OCR tool only. No PDF24 Creator, offline or existing-text editing outcome is established.

Download original OCR PDFDownload extracted PDF textView actual completion screenSee the exact online record

Sejda Online OCR: saved output and evidence

This is Sejda’s separate online OCR tool, not an existing-text edit in its PDF editor. No Sejda Desktop or editor-fidelity outcome is established.

Download original OCR PDFDownload extracted PDF textDownload Sejda’s original text outputDownload Sejda’s original ZIPView actual completion screenSee the exact online record

Method, dates and limitations

Tested 3 October 2026 in a remote Chrome Browser session. Vendor web build numbers, OS build and hardware were not independently established. We rendered the PDFs with Poppler, extracted text with Poppler, PyMuPDF and pypdf, and recorded PyMuPDF search rectangles for each marker. The source and output hashes identify the actual files.

No handwriting, mixed languages, real damaged scans, batches or difficult layouts were tested. This does not measure speed, reliability, privacy, server deletion, accessibility, long-term free limits or editing quality. The full OCR fixture protocol and existing-text editing task remain uncompleted by this limited check.

Read the public test recordDownload SHA-256 artifact manifestInspect extraction and search resultsRead the downloadable report

Exact tools used on 3 October 2026: PDF24 Online OCR and Sejda Online OCR. These run dates do not refresh the catalog’s vendor-source review dates.

Published fixture outcomes

Every relevant edition is shown. “Not run” is an empty evidence slot, not a failure or a vendor feature tick. Paid-edition hands-on tests are not part of the current test program.