This is the FuzzyOcr Frequently Asked Questions (FAQ). Please read it before sending any support requests :)


Question 1:  I've installed FuzzyOcr plugin according to the INSTALL instructions,
             but it doesn't seem to do anything, what can I do?

Answer 1:    Try running FuzzyOcr on the samples provided within this tarball,
             or download them seperately from the download page.
             The archive contains a README file with instructions to test.


Question 2:  I've installed FuzzyOcr plugin according to the INSTALL instructions,
             and I want to see if it is all working correctly.

Answer 2:    See Answer 1.
 
 
Question 3:  I've ran SA on the samples, but FuzzyOcr isn't doing anything.

Answer 3:    First of all, enable the debug mode, setting focr_verbose 2 in the config file.
             Also make sure, that the logfile specified in the config file is writable.
             Then run one of the samples, and then check the logfile for messages indicating
             errors. See the remaining questions if you can't resolve an error message.

Question 4:  My installation is working but I'm still getting image spam, what can I do?

Answer 4:    There are several steps you can try, to get rid of remaining image spam:
                - Save the image, if it is a gif file, analyze wether it is animated/interlaced or normal.
                - On a normal picture, run gocr -i filename
                  Check the output, if it looks garbage only (i.e. even with a bit of approximation, there is no word the plugin could match)
                  then you need to try different settings, try experimenting with the -l setting, if the image is noisy, try the -d parameter,
                  experiment with the values. If you get good results and are getting this kind of spam a lot, then add this setting
                  to your scansets.
                  Also make sure, that you have enough keywords for this kind of spam in your wordlist.
                - If you fail to get a usable result with gocr alone, try involving pnm processors, like pnmnorm, pnmquant or pnminvert.
                  There are no limits in what you can involve in a scanset to get text from a pnm file.
                  You can even use a commercial software, although things like that were never tested.
                  In case you find scansets which generally improve the recognition rate, please send them to the mailing list.
                - If even that fails, you can still add the md5 sum to the md5 database manually, if you are getting this image often.

Question 5:  I'm often getting false positives because mails contain screenshots, what can I do?

Answer 5:    There are some things you can try:
                - Decrease the focr_threshold value to 0.2 or 0.21, that makes the matching more exactly.
                - Check if the false positives are caused only by specific words on your wordlist and remove these error prone words.

Question 6:  I'm using Redhat or a Redhat based distribution and all my gocr results look bad.

Answer 6:    On Redhat based systems, some RPMs/SRPMs are built incorrectly with the parameter "--with-netpbm=no
.
             This is wrong, you need to make sure that you have a gocr build compiled WITH netpbm support.

Question 7:  My gocr segfaults on some pictures.

Answer 7:    Please patch your gocr source with the patch available on my download page and rebuild it.

Question 8:  My giftext segfaults on some pictures.

Answer 8:    Please patch your giftext source with the patch available on my download page and rebuild it.

Question 9:  I'm getting "Failed to open pipe to external programs with pipe command..."

Answer 9:    This indicates a failure in opening the pipe itself, most likely this is caused by a missing binary.

Question 10:  I am using MailScanner and I'm getting "Unexpected error in pipe to external programs...."
              with the graphic tools pipes (like jpegtopnm failing).

Answer 10:    MailScanner by default only passes the first 30kb of the mail to SpamAssassin.
              Sometimes, this causes the image to be truncated in the middle if it is bigger.
              The only way to fix this at the moment is disabling this option in MailScanner (see your documentation).

Question 10:  I'm getting "Unexpected error in pipe to external programs...."

Answer 11:    This indicates a failure in the pipe somewhere, most likely this is caused by a missing binary.
             Also, if a program within the executed chain fails, this will cause such an error.
             If you get this only rarely on some specific mails, then can be caused by extremely broken images.
             To find out which binary fails, get the picture which causes the error, and run the chain of programs
             manually over the picture. You can do this step by step and check for error messages then.
             If you get pictures that cause such errors in the program chain, please send them to me.

Question 12: I'm getting "Skipping scanset "xyz" because of errors, trying next...",
             what does that mean?

Answer 12:   This indicates that the scanset command "xyz" failed, either because a program
             in the scanset was missing, or produced an error. This doesn't need to be your fault,
             especially with the pnmquant scanset, this can happen with some images.
             This is only critical if you get it for every scanset, that most likely indicates that
             your gocr path is wrong or something else is wrong with the gocr binary (check Question 7).