Why OCR Misses Equations, Footnotes, Tables, and Diagrams

OCR turns visible page images into searchable text, but textbooks contain structures that are not ordinary sentences. Equations, footnotes, tables, diagrams, multi-column layouts, and small labels can be recognized incorrectly even when the page looks sharp.

Short answer: Use OCR to find the likely page. Use the visible page or official source text to verify formulas, quotations, names, and numbers.

Why textbook pages are difficult

Text OCR usually assumes a line-by-line reading order. Textbooks regularly break that assumption.

Content	Common OCR failure
Equation	Loses fractions, superscripts, subscripts, or symbol order
Footnote	Inserts it into the main paragraph or skips small type
Table	Reads across the wrong row or merges columns
Diagram	Captures labels without arrows or spatial relationships
Two-column page	Jumps between columns
Chemical notation	Confuses bonds, charges, and element symbols
Code sample	Changes punctuation, indentation, or similar characters
Scanned page	Adds errors from blur, skew, shadows, or compression

An OCR layer can therefore be searchable without being faithful enough to quote.

Equations are two-dimensional

Consider a fraction with a superscript inside the numerator. A reader sees vertical placement and grouping. Ordinary OCR may output a flat sequence with no reliable boundary.

Watch for:

• minus signs changed to hyphens;
• multiplication signs changed to x;
• Greek letters changed to Latin letters;
• superscripts moved into the main line;
• subscripts dropped;
• fraction bars omitted;
• matrix rows read in the wrong order;
• equation numbers mixed into the formula.

Never paste an OCR equation into an assignment without comparing it character by character with the page.

These are representative errors to look for, not predictions about every OCR engine:

Visible notation	Risky OCR result	Why it matters
x²	x2	Exponent becomes a coefficient or adjacent digit
H₂O	H20	Subscript 2 becomes the number 20
5 × 10⁻³	5 x 10-3	Multiplication and exponent structure flatten
−0.05	-0.05 or 0.05	Mathematical minus can change or disappear
25 µg	25 ug	Unit symbol changes and may be misread

Footnotes are small and out of order

Footnotes use smaller text and sit outside the main reading flow. OCR may append them in the middle of a paragraph, place the marker after the wrong sentence, or miss the note entirely.

When citing a footnote, inspect the visible marker, note text, page label, and edition. Keep the footnote number with the quotation.

Tables need row and column structure

Plain text output cannot always preserve a grid. A value can move under the wrong heading even when every character was recognized correctly.

Use the page image to verify:

• column headings;
• units;
• decimal places;
• negative signs;
• merged cells;
• row labels;
• notes below the table.

For data you must reuse, transcribe the smallest necessary section and have a second pass compare it with the image.

Diagram labels are not the diagram

Recognizing "mitochondrion," "input," or "north" does not preserve which arrow points where. OCR does not turn the visual relationship into a complete semantic model.

Keep the diagram image with its caption and figure number. Use a publisher-provided accessible description when available. The W3C's complex-image guidance explains that diagrams, charts, and maps often need both a short identification and a longer description of their structure, values, and relationships. A list of OCR-recognized labels is not an equivalent replacement.

If an accessible description is missing, report the title, edition, figure number, page, and required assistive technology to the publisher or accessibility office.

A practical verification workflow

For material you are permitted to process:

1. Prefer the official PDF or EPUB text layer.
2. Keep the original page image unchanged.
3. Add OCR as a search aid.
4. Search for a distinctive term to locate the page.
5. Verify the visible sentence or symbol before copying.
6. Check one prose page, one equation, one table, and one figure.
7. Record known limitations with the document.

Treat the visible page as the reference and the OCR layer as an index.

Audit four representative pages

Do not judge a whole textbook from one clean paragraph. Use a small sample that covers the page types most likely to fail:

Sample page	What to compare	Pass condition
Ordinary prose	Two sentences, one proper name, one page number	Words and reading order match the page
Equation page	Operators, fractions, superscripts, subscripts, equation number	Every symbol and grouping matches visually
Table page	Two row labels, two headings, four intersection values, units	Each value remains under the correct heading
Figure page	Caption, label text, arrow destinations, legend	Search finds labels; visual relationships are reviewed separately

Record the pages that failed and use OCR search results from those page types only as navigation hints.

Improve recognition before capture

When you own the material or have permission to capture it:

• use the highest readable zoom without clipping;
• keep the page upright;
• use light mode for dark text on a light background when the source allows it;
• wait for fonts and images to finish loading;
• avoid translucent toolbars over text;
• capture at a consistent size;
• test a complex page before a long sequence.

Increasing resolution cannot repair a page that was already blurred, clipped, or half-loaded.

Do not OCR around a copying restriction

OCR is not a permission tool. Do not use it to bypass a reader's disabled copy function, DRM, print allowance, or extraction prohibition.

If you need accessible text, contact the publisher, library, or institution's accessibility office. An approved source file can preserve headings, reading order, math markup, and alt text far better than OCR.

The image-only versus searchable PDF explanation sets realistic expectations for the text layer.

A review note to keep with the file

OCR is provided for search only.
Verify quotations, equations, table values, footnotes,
figure labels, and page references against the visible page.

This simple warning prevents someone else from treating approximate text as a verified transcription.

FAQ

Why does OCR struggle with equations?

Math depends on two-dimensional position and specialized symbols that ordinary text OCR can flatten.

Can OCR understand a diagram?

It may find labels, but it usually does not preserve their visual relationships.

Is OCR good enough for textbook search?

Often for clear prose. Use it to find pages, then verify important details visually.

How can I improve results?

Use the official source when possible, keep captures sharp and upright, wait for loading, and test complex pages early.