Practical guide
Text extraction and visual preservation are different jobs
A PDF can contain embedded text, raster images, vector drawings, and tables positioned as separate text fragments. Basic extraction may recover the words while losing the relationship between a caption and its figure or between a table cell and its column header. DocKernel's browser preview can identify pages with image operations and represent those pages visually inside the Markdown preview. That is useful for inspection, but an embedded data URL is not the same as a clean asset package for a repository or publishing system.
For durable output, decide what must remain editable and what may remain visual. Narrative text should become headings and paragraphs. A simple table can be rebuilt as Markdown when row and column relationships are unambiguous. A dense financial table, chart, form, or diagram may be safer as a referenced image plus a concise caption and surrounding text. Account-backed conversion jobs can expose downloadable assets when the configured processor returns them, letting the Markdown and resource files travel together.
Use this workflow when visual context matters
- Product manuals contain screenshots that must stay next to instructions.
- Reports include charts whose captions and page references need to remain traceable.
- Tables must be reviewed for row, column, merged-cell, and continuation errors.
- A documentation migration needs separate assets instead of large inline base64 image data.
Workflow
Convert visual PDFs without losing the review trail
- 01
Inspect the source
Identify which pages contain meaningful figures or tables and which are mainly decorative.
- 02
Run a browser preview
Use local extraction to review text order and see which pages are represented visually.
- 03
Choose the final representation
Keep simple tables as Markdown; use referenced assets for complex visuals; add captions and page markers.
- 04
Verify the package
Open the Markdown and every asset, check relative paths, compare captions, and confirm no table row or figure was silently dropped.
Conversion example
Example: keep a figure reference understandable
The output does more than insert an image. It keeps the figure name, a descriptive caption, and a page marker so the resource remains meaningful when viewed outside the original PDF.
Source PDF content
Page 6
Figure 2. Request lifecycle
[diagram: upload → queue → convert → download]
Median processing time excludes queue delay.Markdown output
<!-- page: 6 -->
## Request lifecycle
*Figure 2. Upload, queue, conversion, and download stages.*
Median processing time excludes queue delay.Quality benchmark
Images and tables acceptance benchmark
Judge visual documents by whether meaning survives, not by whether every coordinate is reproduced. The correct output format can differ by page type.
| Document profile | Recommended route | Pass criteria |
|---|---|---|
| Screenshot-based user guide | Markdown text plus referenced image assets | Every instruction points to the correct screenshot and all asset paths resolve. |
| Simple comparison table | Reconstructed Markdown table | Headers, row labels, values, and footnotes match the source after spot checks. |
| Dense chart, form, or merged-cell table | Referenced image plus explanatory text | No invented cell structure; caption, page marker, and surrounding interpretation remain available. |
This benchmark is an acceptance matrix, not a claim that every PDF reaches the same accuracy. Test representative files from your own collection before automating a large queue.
Visual conversion limitations
Markdown is intentionally simpler than PDF layout. Some visual information should be represented rather than force-fit into a fragile table or text approximation.
- Inline browser preview images can make Markdown very large because they use data URLs.
- Vector diagrams and charts do not automatically become editable Mermaid or SVG source.
- Merged cells, rotated headers, nested tables, and tables spanning pages require careful manual review.
- Image-only captions and labels may require OCR before they can become searchable text.
Frequently asked questions
PDF images and tables questions
1Does the browser converter extract image files?
The browser preview can render pages that contain images and include visual data in the preview. Standalone downloadable assets belong to the account-backed processor workflow when the configured processor returns them.
2Can every PDF table become a Markdown table?
No. Simple, clearly separated rows and columns are good candidates. Merged cells, spanning headers, wrapped values, and page breaks can make a visual image or structured CSV-style review safer.
3Why is Markdown with embedded images so large?
Browser previews may use base64 data URLs so the result remains self-contained. For production documentation, separate image files with relative links are usually smaller, cacheable, and easier to manage.
4Will charts become editable data?
Not by default. A chart may be preserved as an image and its visible text may be extracted, but recovering the underlying dataset requires a separate chart or table extraction process.