I would like to search in a pdf file for all the pages, each containing several given words in no particular order. For example, I want to find all the pages which contain both "hello" and "world" in no particular order.

I am not sure if pdfgrep can do it.

I am trying to do something similar to how we can search for several words in a book shown in Google Books.



Yes, you can do it with zero-width lookahead assertions, if you use the -P option (which let it use the PCRE engine and perl-like regexps).

$ pdfgrep -Pn '(?=.*process)(?=.*preparation)' ~/Str-Cmp.pdf
8:•     If a preparation process is used, the method used shall be declared.
10:Standard, preparation may be an important part of the ordering process. See Annex C for some examples of
38:padding. The preparation processing could move the original numerals (in order of occurrence) to the very

The above will only works if the two words are on the same line; if the words can occur on separate lines of the same page, the following will do:

$ pdfgrep -Pn '^(?s:(?=.*process)(?=.*preparation))' ~/Str-Cmp.pdf
8:ISO/IEC 14651:2007(E)
9:                                                                                                  ISO/IEC 14651:2007(E)
10:ISO/IEC 14651:2007(E)
12:ISO/IEC 14651:2007(E)

The s flag in (?s: means that . will also match a newline. Notice that that will only print the first line of the page; you can adjust that with the -A option:

$ pdfgrep -A4 -Pn '^(?s:(?=.*process)(?=.*preparation))' ~/Str-Cmp.pdf
8:ISO/IEC 14651:2007(E)
8-•     Any specific internal format for intermediate keys used when comparing, nor for the table used. The use of
8-      numeric keys is not mandated either.
8-•     A context-dependent ordering.
8-•     Any particular preparation of character strings prior to comparison.
9:                                                                                                  ISO/IEC 14651:2007(E)

A crude wrapper script that will print the lines matching any of the patterns from the pages matching all of the patterns in any order:

usage: pdfgrepa [options] files ... -- patterns ...

#! /bin/sh
r1= r2=
for a; do
        if [ "$r2" ]; then
                r1="$r1(?=.*$a)"; r2="$r2|$a"
                case $a in
                --)     r2='(?=^--$)';;
                *)      set -- "$@" "$a";;
pdfgrep -A10000 -Pn "(?s:$r1)" "$@" | grep -P --color "$r2"

$ pdfgrepa ~/Str-Cmp.pdf -i -- obtains process preparation 37- the strings after preparation are identical, and the end result (as the user would normally see it) could be 37- collation process applying the same rules. This kind of indeterminacy is undesirable. 37-one obtains after this preparation the following strings:

  • Assuming the PDF has the text in it, and not bitmaps. Not all PDFs have text content :-( – Stephen Harris Apr 20 at 2:44
  • @Tim No that's not true. It shows the pages containing all the given words. To show the pages containing any of the given words, pdfgrep -P 'process|preparation' would've sufficed. – mosvy Apr 20 at 3:06
  • Thanks. Why does it show the first line? Is it default behaviour or by some option? – Tim Apr 20 at 3:18
  • Because a zero-width assertion --by definition-- has no width, and the matched part of the page will be the empty string before the 1st char from the page (and consequently, the 1st char from the 1st line from the page). If you want the matched words colored, the simplest thing I can think of is to pipe the output to another | egrep --color 'process|preparation'. You can make the whole thing into a function. – mosvy Apr 20 at 3:23
  • Thanks. -A4 only outputs the first five lines in a matching page, not necessarily the lines containing the query words. How can you show the lines containing the query words? – Tim Apr 20 at 4:20
pdfgrep -nP 'hello.{1,99}world|world.{1,99}hello' a.pdf


  • Could you explain why {1,99}? – Tim Apr 20 at 2:49

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.