How to check if PDF is scanned image or contains text in R-CodePudding

I am conducting a structural equation model on several PDF's (>1000) in R.

However, some PDF's are readable and other are scanned, i.e. I need to run them through an OCR-function.

Therefore, I need to find a way to automatically identify which PDF's contains text and which don't. Specifically, I wish to get find a way to return whether a given PDF should be ran through OCR.

Does anyone know of any functions or packages in R that might help do this - I can find a couple of solutions for Python, but haven't been able to identify some in R.

CodePudding user response：

You could use an approach like this (as @danlooo already suggested but I wanted to spell it out):

files <- list.files("/home/johannes/pdfs/",
                    pattern = ".pdf$",
                    full.names = TRUE)

pdfs_l <- lapply(files, function(f) {
  out <- pdftools::pdf_text(f)
  # I set the test to an arbitrary number of characters, it works for me but you want
  # to maybe fine tune it a bit
  contains_text <- nchar(out) > 15
  if (!contains_text) {
    out <- pdftools::pdf_ocr_text(f)
  }
  data.frame(text = out, ocr = !contains_text)
})

pdfs_l |>
  dplyr::bind_rows() |>
  dplyr::mutate(text = trimws(text)) |>
  tibble::as_tibble()
#> # A tibble: 22 × 2
#>    text                                                                    ocr  
#>    <chr>                                                                   <lgl>
#>  1 "TEAM MEMBERS:\n                                                      … FALSE
#>  2 "WS 21/22                                                             … FALSE
#>  3 "WS 21/22                                                             … FALSE
#>  4 "TEAM MEMBERS:\n                                                      … FALSE
#>  5 "TEAM MEMBERS:\n                                                      … FALSE
#>  6 "Key Concepts in Political Communication\n    @Agenda Setting, Priming… FALSE
#>  7 "Key Concepts in Political Communication\n    @Agenda Setting, Priming… FALSE
#>  8 "ELECTIONS AND CAMPAIGNS\n                                            … FALSE
#>  9 ""                                                                      TRUE 
#> 10 ""                                                                      TRUE 
#> # … with 12 more rows

^{Created on 2022-02-10 by the reprex package (v2.0.1)}