How can I detect the encoding/codepage of a text file?

GvS_11492 0
Viewed 312327

In our application, we receive text files (.txt, .csv, etc.) from diverse sources. When reading, these files sometimes contain garbage, because the files where created in a different/unknown codepage.

Is there a way to (automatically) detect the codepage of a text file?

The detectEncodingFromByteOrderMarks, on the StreamReader constructor, works for UTF8 and other unicode marked files, but I'm looking for a way to detect code pages, like ibm850, windows1252.

Thanks for your answers, this is what I've done.

The files we receive are from end-users, they do not have a clue about codepages. The receivers are also end-users, by now this is what they know about codepages: Codepages exist, and are annoying.

Solution:

  • Open the received file in Notepad, look at a garbled piece of text. If somebody is called François or something, with your human intelligence you can guess this.
  • I've created a small app that the user can use to open the file with, and enter a text that user knows it will appear in the file, when the correct codepage is used.
  • Loop through all codepages, and display the ones that give a solution with the user provided text.
  • If more as one codepage pops up, ask the user to specify more text.
globalization encoding text .net c#
21 Answers

You can't detect the codepage, you need to be told it. You can analyse the bytes and guess it, but that can give some bizarre (sometimes amusing) results. I can't find it now, but I'm sure Notepad can be tricked into displaying English text in Chinese.

Anyway, this is what you need to read: The Absolute Minimum Every Software Developer Absolutely, Positively Must Know About Unicode and Character Sets (No Excuses!).

Specifically Joel says:

The Single Most Important Fact About Encodings

If you completely forget everything I just explained, please remember one extremely important fact. It does not make sense to have a string without knowing what encoding it uses. You can no longer stick your head in the sand and pretend that "plain" text is ASCII. There Ain't No Such Thing As Plain Text.

If you have a string, in memory, in a file, or in an email message, you have to know what encoding it is in or you cannot interpret it or display it to users correctly.

JV_53900
user_115580

You can't detect the codepage

This is clearly false. Every web browser has some kind of universal charset detector to deal with pages which have no indication whatsoever of an encoding. Firefox has one. You can download the code and see how it does it. See some documentation here. Basically, it is a heuristic, but one that works really well.

Given a reasonable amount of text, it is even possible to detect the language.

Here's another one I just found using Google:

Shy_96110

I've done something similar in Python. Basically, you need lots of sample data from various encodings, which are broken down by a sliding two-byte window and stored in a dictionary (hash), keyed on byte-pairs providing values of lists of encodings.

Given that dictionary (hash), you take your input text and:

  • if it starts with any BOM character ('\xfe\xff' for UTF-16-BE, '\xff\xfe' for UTF-16-LE, '\xef\xbb\xbf' for UTF-8 etc), I treat it as suggested
  • if not, then take a large enough sample of the text, take all byte-pairs of the sample and choose the encoding that is the least common suggested from the dictionary.

If you've also sampled UTF encoded texts that do not start with any BOM, the second step will cover those that slipped from the first step.

So far, it works for me (the sample data and subsequent input data are subtitles in various languages) with diminishing error rates.

user_68990

The tool "uchardet" does this well using character frequency distribution models for each charset. Larger files and more "typical" files have more confidence (obviously).

On ubuntu, you just apt-get install uchardet.

On other systems, get the source, usage & docs here: https://github.com/BYVoid/uchardet

user_6270420

The StreamReader class's constructor takes a 'detect encoding' parameter.

user_155410

Got the same problem but didn't found a good solution yet for detecting it automatically . Now im using PsPad (www.pspad.com) for that ;) Works fine

DubCee_58950

Since it basically comes down to heuristics, it may help to use the encoding of previously received files from the same source as a first hint.

Most people (or applications) do stuff in pretty much the same order every time, often on the same machine, so its quite likely that when Bob creates a .csv file and sends it to Mary it'll always be using Windows-1252 or whatever his machine defaults to.

Where possible a bit of customer training never hurts either :-)

devstuff_413210
user_44483530

try and install the perl module Text::Unaccent::PurePerl by typing cpanm Text::Unaccent this generates a build.log file that displays as chinese in some applications as english in others cpanm is the initial text a plausible attempt should you be lucky enough to have spaces in the language is to compare the distribution frequency of words via a statistical test

user_20379860

10Y (!) had passed since this was asked, and still I see no mention of MS's good, non-GPL'ed solution: IMultiLanguage2 API.

Most libraries already mentioned are based on Mozilla's UDE - and it seems reasonable that browsers have already tackled similar problems. I don't know what is chrome's solution, but since IE 5.0 MS have released theirs, and it is:

  1. Free of GPL-and-the-like licensing issues,
  2. Backed and maintained probably forever,
  3. Gives rich output - all valid candidates for encoding/codepages along with confidence scores,
  4. Surprisingly easy to use (it is a single function call).

It is a native COM call, but here's some very nice work by Carsten Zeumer, that handles the interop mess for .net usage. There are some others around, but by and large this library doesn't get the attention it deserves.

user_897060
Related