OCR Explained: How to Convert Scanned PDFs to Editable Text

You receive a PDF, open it, and try to copy a paragraph, but nothing gets selected. You search for a specific word and the results come back empty. The file looks perfectly normal, but it behaves like a picture rather than a document. What you are holding is a scanned PDF, and the only reliable way to make it searchable, editable, and reusable is Optical Character Recognition, commonly known as OCR.

This guide explains what OCR actually does, when you need it, and which approach gives the best results for different situations. Whether you want a quick online conversion, a repeatable command-line workflow, or professional-grade accuracy for archival documents, the steps below will get you from a flat scan to clean, editable text.

What Is OCR and How Does It Work

OCR is a technology that analyzes images of text and converts them into actual character data that computers can read, edit, and search. When you run OCR on a scanned document, the software does not magically see letters. Instead, it performs a sequence of steps that simulate how a human reader would process a page.

The first stage is image preprocessing. The engine straightens skewed pages, removes noise and speckles, normalizes contrast, and binarizes the image so text becomes pure black on pure white. The second stage is layout analysis. The software identifies text regions, columns, headers, footers, tables, and images, and determines the reading order. The third stage is character recognition itself. Classical engines match each glyph against a database of known character shapes, while modern engines such as Tesseract 5 and ABBYY FineReader rely on deep neural networks, particularly LSTM and transformer models, that recognize entire lines of text in context.

The final stage is post-processing, where dictionaries, language models, and grammar rules correct ambiguous characters. For example, the engine decides whether "rn" should actually be "m", or whether "0" should be "O", based on the surrounding words. This is why OCR accuracy is heavily dependent on language selection and training data quality.

Native PDFs vs Scanned PDFs

Before running OCR, you need to know which type of PDF you have, because the treatment is completely different.

A native PDF is one generated from a digital source such as Word, Google Docs, LaTeX, or a design tool. Its text is stored as actual character data with font information. You can select, copy, and search it without any special processing. For these files, you do not need OCR at all, and converting them to other formats is a direct operation. See our PDF vs DOCX guide for a deeper comparison of native document formats.

A scanned PDF is produced when a physical page is photographed or scanned. The PDF wraps one image per page, and there is no text layer at all. Selection tools do nothing. Search returns zero results. File size is usually larger than a native PDF of similar length. This is the category that requires OCR.

A third hybrid category exists: image-only PDFs that have been partially OCR'd, where some pages have a text layer and others do not. These files often cause the most confusion because search works inconsistently. Running a full OCR pass on the entire document resolves the issue.

A quick way to diagnose your file is to open it and press Ctrl+A or Cmd+A. If text highlights in blue selection, it is native. If only an image highlights, you need OCR.

Preparing Documents for Accurate OCR

OCR accuracy is determined at the moment of scanning, not at the moment of processing. A poor scan will produce poor results no matter which engine you use. If you still control the source document, invest a few minutes in getting the scan right.

Resolution is the single most important factor. Aim for 300 DPI for standard printed text at typical body sizes such as 10 to 12 points. For small print, footnotes, or legal fine print, move up to 400 or 600 DPI. Going beyond 600 DPI rarely improves accuracy and dramatically increases file size and processing time.

Color mode matters more than people expect. For typical text documents, scan in grayscale rather than full color. Color scans introduce chromatic noise that confuses the binarization stage. Pure black and white scans can be acceptable for clean modern prints, but they destroy information in faded or low-contrast originals.

Contrast and brightness should be adjusted so that ink is dark and paper is bright without pushing either to extremes. If your scanner has an automatic contrast setting, test a page before committing to a batch. Lighting matters if you are photographing pages with a phone. Use diffuse, even lighting, avoid shadows from your hand or the camera, and keep the camera parallel to the page.

Finally, straighten pages before scanning. Even a two-degree skew hurts recognition of small fonts. Most scanner software can auto-deskew, but you can also let the OCR engine do this during preprocessing.

Comparing the Main OCR Approaches

There are five mainstream ways to OCR a scanned PDF, each suited to different needs. The table below summarizes how they compare.

The accuracy ranges assume reasonably clean 300 DPI scans in a single common language. Degraded documents, handwriting, and multi-script pages produce lower results across every tool.

Step-by-Step Methods

Method 1: Online PDF to Text Converters

The fastest path for a single document is an online converter. Upload your scanned PDF, the service runs OCR in the cloud, and you download the result. Most services let you export as plain text, Word, or a searchable PDF. Use PDF to TXT for raw extracted text, or PDF to DOCX if you want to preserve layout and continue editing in Word. For scans that contain tables, PDF to XLSX often gives better structured output than a Word conversion. The general mechanics of cloud-based conversion are covered in how file conversion works.

Method 2: Adobe Acrobat Pro

Open the PDF in Acrobat Pro, choose Tools, then Scan and OCR, and click Recognize Text, In This File. Acrobat detects the document language automatically, but you can override it. Once processing finishes, the text layer is embedded in the original PDF, so the file looks identical but is now fully searchable. To extract the text itself, use File, Export To, and choose Microsoft Word or Rich Text Format. Acrobat's Action Wizard can batch-process entire folders, which is useful for digitization projects.

Method 3: Tesseract on the Command Line

Tesseract is the gold standard open-source OCR engine and a great choice when you want to automate. After installing Tesseract and a PDF utility such as Poppler, the basic workflow is to split the PDF into images and then run OCR on each image.

pdftoppm -r 300 input.pdf page -png
for img in page-*.png; do
  tesseract "$img" "${img%.png}" -l eng
done
cat page-*.txt > output.txt

To produce a searchable PDF directly, use the pdf output mode:

tesseract input.tiff output -l eng pdf

For multi-page TIFFs created from scanners, Tesseract handles the entire file in one call. Combine this with a shell loop and you can OCR thousands of files unattended. Related image-to-PDF workflows such as JPG to PDF, PNG to PDF, and TIFF to PDF are useful when you need to consolidate scans before OCR.

Method 4: Google Drive

Upload the scanned PDF to Google Drive, right-click it, and choose Open with, Google Docs. Drive performs OCR during the conversion and opens a new Doc containing the recognized text, usually with the original page images above each text block. You can then copy the text, save as Word, or download as plain text. Accuracy is good for clean modern documents and particularly strong for languages with ample Google training data. Formatting fidelity is weaker than Acrobat or ABBYY, so treat Drive as a text-extraction tool rather than a layout-preservation tool.

Method 5: ABBYY FineReader

ABBYY FineReader is the benchmark for commercial OCR. Its strengths are layout reconstruction, complex table recognition, multi-column magazines, and low-quality archival materials. Open the PDF, select your languages (you can choose several at once), review the automatically detected zones, and export to Word, Excel, searchable PDF, or EPUB. If your work involves historical documents, legal contracts, or publications that must match the original layout precisely, FineReader usually justifies its price. For everyday documents, open-source and built-in options are adequate.

Handling Multi-Language Documents and Handwriting

Multi-language OCR is a common stumbling block. If your document mixes English and French, Tesseract accepts multiple language codes at once:

tesseract input.png output -l eng+fra

Acrobat and ABBYY allow multi-language selection in their document properties panel. Google Drive auto-detects but performs best when the dominant language is obvious. For right-to-left scripts such as Arabic and Hebrew, ensure your engine has the correct language pack installed; otherwise, you will get reversed or garbled output.

Handwriting is a different problem. Traditional OCR engines, which were built for printed typefaces, perform poorly on cursive or mixed handwriting. For handwriting specifically, look at specialized ICR (Intelligent Character Recognition) services, Google Cloud Vision's DOCUMENT_TEXT_DETECTION mode, or Microsoft Azure's Read API. Even these top out around 85 percent accuracy on neat block letters and fall below 70 percent on cursive, so always plan on a human review pass for critical handwritten content.

Preserving Document Layout

Raw OCR gives you text, but many workflows need the visual structure preserved too. Preservation quality depends heavily on the tool.

For structured documents such as invoices, reports, and academic papers, export to DOCX rather than TXT. Acrobat and ABBYY reconstruct columns, headers, and paragraph breaks reasonably well. For tables, ABBYY and specialized table-extraction pipelines are much stronger than generic tools; see our guide on how to convert PDF tables to Excel for deeper coverage.

If formatting fidelity matters more than editability, choose the searchable PDF output option. This keeps the original page image as the visible layer and adds a hidden text layer underneath. The file looks identical to the scan, but you can search and select the text. This is the best format for legal discovery, compliance archives, and any situation where the visual artifact must remain unchanged. When moving between Word and PDF in either direction, our convert Word to PDF formatting tips guide covers the common layout pitfalls.

Batch OCR for Large Document Sets

If you have hundreds or thousands of files, manual processing is not viable. Build a repeatable pipeline instead.

With Tesseract, combine shell scripting and a naming convention. Place all source scans in an input folder, iterate over them, and write results to a parallel output folder. Use parallel processing with tools like GNU parallel to cut total time on multi-core machines. Log failures to a separate file so you can retry them without rerunning successful conversions.

Acrobat's Action Wizard provides a graphical alternative. Create an action that applies Recognize Text followed by an export step, and point it at a folder. Acrobat will churn through every PDF and deliver results to the output location you specify.

For enterprise workloads, cloud OCR APIs such as Google Cloud Vision, AWS Textract, and Azure Document Intelligence are designed specifically for scale, with per-page pricing and parallel processing built in. These services also return structured JSON with bounding boxes and confidence scores, which is essential if you need to validate results programmatically.

Tips for Maximizing Accuracy

Start with the best possible source. If you can rescan, do it at 300 to 400 DPI grayscale. If you cannot, run image enhancement before OCR. Tools like ImageMagick can deskew, denoise, and increase contrast in a single command.

Always specify the correct language. Relying on auto-detection works for common languages but often fails on mixed or unusual scripts. When in doubt, specify manually.

Use confidence scores if your tool exposes them. Tesseract, Google Vision, and Azure all return a confidence value per word. Flag anything below 80 percent for human review rather than trusting the full output blindly.

Proofread critical content. Even 99 percent accuracy means one error every hundred characters, which adds up fast in long documents. For contracts, medical records, or financial data, a human review pass is non-negotiable regardless of which engine you use.

When OCR Fails and What to Do About It

OCR has clear limits, and knowing them saves time. Very low resolution scans below 150 DPI rarely recover well even after upscaling. Documents with heavy bleed-through from the reverse side introduce ghost characters that confuse the recognizer. Unusual fonts, especially decorative or highly stylized display types, fall outside the training data of most engines and produce garbled results.

Handwritten notes, as mentioned earlier, benefit from specialized ICR rather than general OCR. Pages with dense mathematical notation or chemical formulas need a math-aware engine such as Mathpix; Tesseract will mangle subscripts, superscripts, and symbols.

If standard OCR is failing, try these recovery steps in order. First, rescan at higher resolution if the original is available. Second, preprocess with contrast enhancement and deskewing. Third, try a different engine; ABBYY often rescues documents that Tesseract cannot read, and vice versa. Fourth, consider splitting the work: OCR what you can automatically, and manually transcribe the rest.