# Check Images, Captions, and References After Scanned PDF Conversion

Verify that converted PDF illustrations remain connected to their captions, figure numbers, and surrounding text, and diagnose missing images.

An illustration arriving in the output folder is only the first check. The reader must still know what it shows, which caption belongs to it, and why the body text refers to it. A diagram separated from its labels can be as misleading as a missing diagram.

Review scanned PDF image and caption extraction as a set of relationships. Keep the original page open while checking the converted document and any accompanying resources.

## Inventory the figures you need

Choose a chapter or section with several kinds of visual content: a photograph, a diagram, and a chart if the source has them. Note each figure number, its original page, and the first words of its caption.

Open those figures in the result. Check that the edges are present and readable: axis labels, legends, scale bars, arrows, and panel labels are often near the boundary. An image cropped just slightly too tightly can lose the part that explains its meaning.

Distinguish image preservation from text recognition. Keeping a chart as an image does not mean that every label inside it becomes selectable or translated. If the labels remain in the source language, record that explicitly rather than assuming the entire illustration was localized.

## Match each caption to its image

Read the figure number and full caption in the scan, then locate the corresponding text in the output. A multi-line caption can be split into separate paragraphs, mixed with neighboring body text, or attached to another image.

For a figure with panels A, B, and C, confirm that the caption still explains all three. Watch for a legend printed beside the image rather than beneath it. The relevant information may not fit a simple “image followed by caption” pattern.

Check punctuation and note markers inside captions as carefully as the main text. A missing “not” or a lost significance marker can alter a chart's interpretation even though the picture itself is unchanged.

## Follow body references through the document

Find a sentence that says “Figure 2 shows…” and trace it to the correct figure. Then return to read the paragraph immediately before and after the figure's new location.

In reflowed output, the illustration need not sit at the same physical position as in the PDF. It should still appear in a sensible place and remain identifiable from its reference. A caption inserted halfway through an unrelated sentence is a reading-order error, not simply a new layout.

When numbering restarts by chapter or appendix, include that context in your checks. Two different “Figure 1” labels may both be correct; renumbering one casually can break the author's references.

## Diagnose a missing image before replacing it

If the document shows a broken-image placeholder, inspect the Markdown link and its resource location. The image may exist but have moved when you copied only the `.md` file. A relative path is resolved from the Markdown document, not from wherever you originally downloaded it.

If there is no link, search the extracted resources and compare with the source. Distinguish a missing file, an omitted link, a crop problem, and an unsupported display format. They require different remedies. The [Markdown image-assets guide](https://pdfcraft.ai/en/blog/pdf-markdown-image-assets/) covers copying and sharing a result without breaking its paths.

Do not substitute a generated illustration for a missing source diagram. That would create new information rather than recover the document. If recovery is not practical, keep a clearly identified reference to the original page.

## Check the file where you will actually use it

For editable content, upload the PDF to [PDF to Markdown](https://pdfcraft.ai/en/pdf-craft/pdf-to-markdown/), select Markdown, and download the completed result. PDF Craft processes document content and structure, but scan quality and complex layouts still affect extraction. Inspect the actual downloaded file rather than assuming a fixed package structure or automatic image hosting.

Open the result in your intended editor or reader. Try one figure at a larger zoom, follow its reference, and verify the caption. If your goal is research notes, preserve the original PDF and figure page when following the [research Markdown workflow](https://pdfcraft.ai/en/blog/pdf-to-markdown-for-research/).

A useful acceptance record names the figure, caption, source location, and any unresolved defect. It does not reduce the result to an image count: one correctly connected figure is more informative than three files with ambiguous names.