Skip to contents

Extracting text and fonts

Source: vignettes/text-extraction.Rmd
text-extraction.Rmd

pdfium exposes text at the level of text runs — the contiguous runs of glyphs that share a font, size, and rendering state. Each run carries its bounding box, the Unicode content, and font metadata. Where pdftools::pdf_doc_text() returns plain strings, pdfium lets you connect each character to where it sits on the page and what font drew it.

library(pdfium)
fixture <- system.file("extdata", "fixtures", "unicode.pdf",
  package = "pdfium"
)
doc <- pdf_doc_open(fixture)
page <- pdf_page_load(doc, 1L)

Whole-page extraction

pdf_text_runs(page) returns one row per text-typed page object on the page, with content + position + font in a single tibble:

runs <- pdf_text_runs(page)
runs
#> # A tibble: 5 × 13
#>   obj_index bounds_left bounds_bottom bounds_right bounds_top font_size text 
#>       <int>       <dbl>         <dbl>        <dbl>      <dbl>     <dbl> <chr>
#> 1         2        130.          171.         158.       180.         1 Hello
#> 2         3        129.          137.         158.       146.         1 world
#> 3         4        126.          103.         138.       114.         1 pd   
#> 4         5        140.          105.         145.       114.         1 fi   
#> 5         6        146.          105.         162.       112.         1 um   
#> # ℹ 6 more variables: font_base_name <chr>, font_family <chr>,
#> #   font_weight <int>, font_italic_angle <int>, font_is_embedded <lgl>,
#> #   font_flags <int>

Columns (use names(runs) for the canonical list — the names below match the schema PDFium reports today):

  • object_index — 1-based page-object index. Matches the index in pdf_page_objects(page).
  • bbox_left, bbox_bottom, bbox_right, bbox_top — bounding box in PDF user space (points, origin bottom-left).
  • font_size_pt — the font size as PDFium reports it (in user-space points, before the text matrix is applied).
  • text — Unicode content of the run, UTF-8 encoded.
  • font_base_name — BaseFont from the font dictionary.
  • font_family — FontFamily if present.
  • font_weight — integer 100..900 (CSS-style; 400 ≈ regular, 700 ≈ bold).
  • font_italic_angle — italic angle in degrees, 0 for upright fonts.
  • font_is_embedded — logical; TRUE if the font program is embedded.
  • font_flags — PDF font-descriptor /Flags bitmask (fixed-pitch, serif, symbolic, …).

Per-object access

When you already have a specific text-typed pdfium_obj in hand — typically from filtering pdf_page_objects() — three accessors give you the same data piecewise:

texts <- Filter(function(o) o$type == "text", pdf_page_objects(page))
length(texts)
#> [1] 5

pdf_text_content(texts[[1L]]) # the Unicode string
#> [1] "Hello"
pdf_text_font_size(texts[[1L]]) # font size in user-space points
#> [1] 1
str(pdf_text_font(texts[[1L]])) # named list of font metadata
#> List of 6
#>  $ font_base_name   : chr "NimbusSans-Regular"
#>  $ font_family      : chr "Nimbus Sans"
#>  $ font_weight      : int 400
#>  $ font_italic_angle: int 0
#>  $ font_is_embedded : logi TRUE
#>  $ font_flags       : int 4

pdf_text_runs() is essentially a batched form of these three accessors joined with pdf_obj_bounds(), returned as a single tibble.

Unicode and font notes

UTF-16LE comes through correctly for every script PDFium knows how to decode — including CJK ideographs, Latin diacritics, and characters outside the Basic Multilingual Plane that need surrogate pairs:

Encoding(pdf_text_content(texts[[1L]]))
#> [1] "unknown"
nchar(pdf_text_content(texts[[1L]]))
#> [1] 5

A few caveats worth knowing about:

  • Glyph IDs vs Unicode. If a PDF uses a non-standard font without a /ToUnicode CMap, PDFium can’t always map glyph IDs back to Unicode. In those cases pdf_text_content() returns the best-effort decoding; some characters may come through as private-use codepoints or as ?. This is a property of the source PDF, not of pdfium.
  • Text matrix. font_size_pt is the font size that PDFium reports. The effective glyph height on the page is font_size_pt * abs(matrix$d) where matrix comes from pdf_obj_matrix(text_obj).

Combining text and path data

The pdf_extract_paths() one-call helper attaches pdf_text_runs() output as a text_runs attribute on its result, so downstream consumers that want to label paths with nearby text have everything they need in one object. See the extracting-paths vignette for the schema.

Cleanup