GNU ELPA - doc-toc

For PDFs without TOC pages, with a very complicated TOC (i.e. that require much cleanup work) or with headlines well fitted for automatic extraction (you will have to decide for yourself by trying it), consider to use the pdf.tocgen functionality described below.

Otherwise, start with opening some pdf or djvu file in Emacs (pdf-tools and djvu package recommended). Find the pagenumbers for the TOC. Then type M-x doc-toc-extract-pages, or M-x doc-toc-extract-pages-ocr if doc has no text layer or the text layer is bad, and answer the subsequent prompts by entering the pagenumbers for the first and the last page each followed by RET. For PDF extraction with OCR, currently it is required to view all contents pages once before extraction (doc-toc uses the cached file data). For TOC's that are formatted as two columns per page, prepend the doc-toc-extract-pages-ocr command with two universal arguments. Then after you are asked for the start and finish pagenumbers, a third question asks you to set the tesseract psm code. For the double column layout it is best (as far as I know) to use psm code 1. Also the languages used for tesseract OCR can be customized via the doc-toc-ocr-languages variable.

A buffer with the, somewhat cleaned up, extracted text will open in TOC-cleanup mode. Prefix command with the universal argument (C-u) to omit cleanup and get the raw text. If the extracted text is of too low quality you either can hack/extend the doc-toc-extract-pages-ocr definition, or alternatively you can try to extract the text with the python document-contents-extractor script, which is more configurable (you are also welcome to hack on and improve that script). For this the tesseract documentation might be useful.

If you merely want to extract text without further processing then you can use the command doc-toc-extract-only.

In this mode you can further cleanup the contents to create a list where each line has the structure:

TITLE (SOME) PAGENUMBER

There can be any number of spaces between TITLE and PAGE. The correct pagenumbers can be edited in the next step. A document outline supports different levels and levels are automatically assigned in order of increasing number of preceding spaces, i.e. the lines with the least amount of preceding spaces are assigned level 0 etc., and lines with equal number of spaces get assigned the same levels.

Contents   1
Chapter 1      2
 Section 1 3
  Section 1.1     4
Chapter 2      5

There are some handy functions to assist in the cleanup. C-c C-j jumps automatically to the next line not ending with a number and joins it with the next line. If the indentation structure of the different lines does not correspond with the levels, then the levels can be set automatically from the number of seperators in the indices with M-x doc-toc-cleanup-set-level-by-index. The default seperator is a . but a different seperator can be entered by preceding the function invocation with the universal argument (C-u). Some documents contain a structure like

1 Chapter 1    1
Section 1      2

Here the indentation can be set with M-x replace-regexp ^[^0-9] -> \& (where there is a space character before the \&).

Type C-c C-c when finished

This mode provides the functionality for easy adjustment of pagenmumbers. The buffer can be navigated with the arrow up/down keys. The left and right arrow keys will shift down/up all the page numbers from the current line and below (combine with SHIFT for setting individual pagenumbers).

The TAB key jumps to the pagenumber of the current line, while C-right/C-left will shift all remaining page numbers up/down while jumping/scrolling to the line its page in the document window. Because the numbering of scanned books often breaks at sections of a certain level, C-j will let jo jump quickly to the next entry of a certain level (e.g. you can quickly check if the page numbers of all level 0 sections correspond to the page numbers in the document). The S-up/S-down in the tablist window will just scroll page up/down in the document window and, C-up/C-down will scroll smoothly in that window.

If you discover some small error in some field, then you put the cursor on that field and press r to correct the text in that field.

Type C-c C-c when done.

The text of this buffer should have the right structure for adding the contents to (for pdf's a copy of) the original document. Final adjustments can be done but should not be necessary. Type C-c C-c for adding the contents to the document.

By default, the TOC is simply added to the original file. (ONLY FOR PDF's, if the (customizable) variable doc-toc-replace-original-file is nil, then the TOC is added to a copy of the original pdf file with the path as defined by the variable doc-toc-destination-file-name. Either a relative path to the original file directory or an absolute path can be given.)

Sometimes the pdfoutline/djvused application is not able to add the TOC to the document. In that case you can either debug the problem by copying the used terminal command from the *messages* buffer and run it manually in the document's folder iside the terminal, or you can delete the outline source buffer and run doc-toc--tablist-to-handyoutliner from the tablist buffer to get an outline source file that can be used with HandyOutliner (unfortunately the handyoutliner command does not take arguments, but if you customize the doc-toc-handyoutliner-path and doc-toc-file-browser-command variables, then Emacs will try to open HandyOutliner and the file browser so that you can drag the file contents.txt directly into HandyOutliner).