{{ summaryTitle }}
{{ summaryValue }}

{{ summaryLine }}

Source{{ sourceBadge }} Route{{ routeLabel }} Pages{{ pageBadge }}
Ignored {{ ignoredFiles }} extra file{{ ignoredFiles === 1 ? '' : 's' }}; this workflow processes one PDF at a time.
{{ dropzoneTitle }}
{{ dropzoneHelp }}
{{ formatBytes(selectedFileMeta.size) }}
The sample has two text-native pages and follows the same PDF.js extraction path.
Auto reports the resolved route for every page; pin OCR when the embedded layer is incomplete or incorrect.
This setting affects raster OCR only; embedded PDF text keeps its existing characters.
Examples: 1, all, 2-4, or 1,3,8-10.
Use 200 DPI for ordinary scans, 300 DPI for small print, or 144 DPI for a quick check.
DPI
Page headings are easiest to audit when a result spans several pages.
{{ progressPercent }}%

{{ progressLabel }}

Automatic page is the neutral default and applies only to raster OCR.
The default guard is 8 pages; raise it only after reviewing range and DPI.
pages
Off is neutral for normal prose.
Leave blank for ordinary PDFs.
{{ extractedText }}
PageRouteConfidenceWordsCharactersRuntimeCopy
{{ row.page }}{{ row.route }}{{ row.confidence }}{{ row.words }}{{ row.characters }}{{ row.runtime }}
PDF Compressor Compress a PDF locally in your browser, compare structural and production estimates, and catch encryption or image-quality risks before download.
PDF Page Extractor Extract selected PDF pages in your browser, keep a custom order or remove repeats, and verify every output page against its source.
PDF Repair Tool Inspect and repair a damaged PDF in your browser, compare structural warnings, and create a normalized copy when its page tree can be read.
PDF to File Converter Convert a PDF locally into editable text or page-faithful images with page-range checks, document handoffs and no source-file upload.

Two PDFs can look identical while storing words in completely different ways. A document exported from an office application often contains characters that can be selected and searched. A scan may contain only page images, so the visible words have to be recognized from pixels before they can be copied, indexed, or translated.

Optical character recognition (OCR) converts printed characters in an image into text. It is useful for scanned letters, receipts, signed forms, archived reports, and mixed packets where some pages have native text and others do not. Reading an existing text layer is faster and usually more faithful, but a text layer can also be incomplete, out of order, or left behind by an earlier low-quality OCR pass.

Embedded text
Characters already stored in the PDF and read without raster recognition.
Raster OCR
Recognition performed after a page is rendered to a pixel canvas.
Page segmentation
The assumed arrangement of text, such as an automatic page, single column, uniform block, or sparse text.
Confidence
A recognition-engine clue for an OCR page, not a guarantee that every character is correct.

Scan quality often matters more than a single setting. Small print benefits from more pixels, while skew, blur, uneven lighting, compression artifacts, handwriting, tables, stamps, and multi-column layouts can reduce accuracy or scramble reading order. Higher resolution also increases memory use and processing time, so a small representative page range is the safest place to tune language and layout.

Extracted text is a draft for verification whenever an error could change meaning. Names, account numbers, dates, totals, decimal points, legal clauses, and table cells should be compared with the page image before the text is used for filing, search, quotation, or data entry.

How to Use This Tool:

Test a few representative pages before increasing the range, especially for a long scanned document.

  1. Drop or browse one PDF file. If it is protected, enter the PDF password in Advanced and run the inspection again.
  2. Choose Auto detect text or OCR to use a usable embedded layer first, or OCR rendered pages to rasterize every selected page.
  3. Select the Printed language and enter Pages as all, one page, a range such as 2-4, or a comma-separated selection such as 1,3,8-10.
  4. Set OCR resolution. Start at 200 DPI, use 300 DPI for small print, or reduce it for a faster diagnostic pass.
  5. Choose the output Text layout. In Advanced, change Page segmentation or Preserve OCR spacing only when the page structure calls for it.
  6. Keep the selection within the Browser page guard, then extract the text. Narrow the range or lower the DPI if the page guard or canvas-size warning stops the run.
  7. Compare the extracted text with the original pages and use the per-page route, word count, blank-page flag, and OCR confidence to decide where proofreading is most important.

Interpreting Results:

Embedded text means at least 24 normalized characters were found in the existing page text layer, so OCR was skipped in Auto mode. This is efficient but does not prove that the layer is complete or in the right reading order. Select OCR-only when the visible page and embedded text disagree.

Raster OCR pages show the recognition confidence returned for that page. Use lower confidence, an empty result, or an unexpectedly small word count as a review cue. A high value can still conceal a wrong digit, punctuation mark, name, column order, or table association.

The overall average mixes OCR confidence with a value of 100 assigned to embedded-text pages for aggregation. It is therefore a route summary, not a document-wide OCR accuracy measurement. Read the per-page route before comparing averages across documents.

Technical Details:

Text recovery is decided page by page. Auto mode first reads text items from the PDF layer, groups nearby items into lines, normalizes whitespace, and accepts that route when the result contains at least 24 characters. Otherwise the page is rendered at the selected DPI and sent through OCR with the selected language, segmentation mode, and spacing policy.

Transformation Core

StageEmbedded-text routeOCR route
SourceText items already stored on the page.Rendered page pixels.
Route conditionAuto mode finds at least 24 normalized characters.OCR-only mode, or Auto mode finds fewer than 24 characters.
Layout controlExisting character positions are grouped into nearby lines.Page segmentation mode 3, 4, 6, or 11 guides recognition.
Language controlNot applied.English, Spanish, French, German, Italian, Portuguese, Dutch, or supported English combinations.
Confidence displayReported as native text.Recognition confidence from 0% to 100%.

The selected page expression expands in the order entered, while duplicates are ignored with a warning. A range must count upward, start at page 1 or later, and end within the inspected PDF. The page guard accepts 1 to 50 selected pages. OCR resolution is bounded from 120 to 300 DPI.

Formula Core

Per-page word counts split normalized non-empty text on whitespace. Totals are exact sums, and average confidence is the arithmetic mean of the page values used by the recovery model.

Wtotal=∑i=1nWi C¯=∑i=1nCin

Here, n is the number of processed pages, W is a word count, and C is a page confidence percentage. Embedded-text pages contribute 100 to the average; OCR pages contribute the returned recognition value. Blank pages are those whose normalized text is empty.

Raster dimensions scale from PDF points using DPI divided by 72. A page is rejected before OCR when width multiplied by height exceeds 64,000,000 pixels. This protects browser memory but is not an exact prediction of peak memory use.

Limitations, Privacy, and Accuracy Notes:

The selected PDF and password are read in the browser and are not sent to a tool-specific document backend. PDF parsing, OCR code, and OCR language data load from external distribution services when needed, so the initial run may require network access.

  • OCR does not preserve page geometry, fonts, images, tables, or semantic document structure in the text output.
  • Handwriting, rotated pages, complex mathematics, decorative fonts, and dense tables can produce poor text even at high confidence.
  • Auto mode can accept an incomplete embedded layer once it reaches 24 characters; use OCR-only when visual content is missing.
  • The password is used to open the PDF locally and is not placed in the address bar or exported text.

Worked Examples:

Mixed office export and scanned signature page

A five-page agreement has native text on pages 1 to 4 and a scanned signature on page 5. Auto mode reads the first four pages directly and sends only page 5 through OCR. The page ledger exposes the mixed routes, so the signature-page names and dates can receive focused proofreading.

Small print exceeds the canvas guard

A large-format page at 300 DPI would exceed 64 million pixels and stops before recognition. Reducing the DPI or selecting fewer pages lowers the raster workload. After extraction, compare small amounts and punctuation with the original rather than assuming the lower-resolution result is exact.

References: