Ripping subtitles using OCR, from .AVI files?

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • ingolf
    Junior Member
    Junior Member
    • Jul 2002
    • 10

    Ripping subtitles using OCR, from .AVI files?

    Hi

    I need to rip subtitles from an AVI file i have recorded from TV, using my tv tuner card.

    I have tried with Subrip v1.17.1, but it only allows to open .VOB files, and when i try to open a image sequence it drops almost all the frames.

    The subtitles appear in a fixed position on the frame, and i have tried to cut out just the subtitle part, using vdub and nulltransform, resulting in bitmap files, dimensions 768x82. If i open this sequence, subrip complains: "134847 skipped, too short"

    The subtitle part does not contain any 'junk' in the background, its completely black, since the movie i recorded has black borders in top and bottom.

    Is there any other program that can do as subrip does, just using avi files, or plain bitmap sequences? Could it be done using avisynth?

    Help appreciated

    Regards
    Ingolf
  • ingolf
    Junior Member
    Junior Member
    • Jul 2002
    • 10

    #2
    After a lot of trial and error, i found out that its possible to get SubRip to recognize the subtitles, as bitmaps.

    I needed to add the filter 'Thresold' in vdub, and export the frames as .BMP files. The Thresold filter ensures that there is only 2 colors in the bitmap, eventhou they are saved as 24bpp.

    Anyway.. i got SubRip working on the images, but after telling SubRip how the letter 'H' looks like, for the 200th time.. i kinda gave it up.

    Seems that SubRip must have an 100% identical frame to work properly. No fuzzylogic/neural network in this ocr engine...

    The size of the pictures that i have exported from vdub is 192x32 pixels

    I have tried adding tons of filters to get the text stand absolutely still, and with minimal flicker, like smoother, denoise, dyn. noise reduction, temporal cleaner, proximity locker, etc etc ... but all to no avail..

    Wonder if such a filter exists for vdub, that will look ahead 1 sec of frames(in my case thats 25 frames), and lock all pixels that have not moved by 2-16 pixels... or something like that.. a real hard scrubber is needed
    Last edited by ingolf; 16 Oct 2003, 02:43 AM.

    Comment

    • Qyd
      Digital Video Technician
      Digital Video Technician
      • Oct 2002
      • 433

      #3
      Seems that SubRip must have an 100% identical frame to work properly. No fuzzylogic/neural network in this ocr engine...
      you're right about that. Subrip is intended for dvd sbtitle streams, that generaly have 100% identical bitmaps for every letter... so as far as ocr'ing from an avi file, i guess you went farther than anyone.
      If you capture is a movie, you could find tons of sites that offer subtitles, if not, i guess it's not worth the effort...

      Comment

      Working...