View Single Post
Old 14th May 2009, 17:55   #378  |  Link
0xdeadbeef
Author of BDSup2Sub
 
Join Date: Jun 2003
Posts: 478
Quote:
Originally Posted by Arshad07 View Post
But still it works, i just have to change it to NTSC in BDS2S....and the texts become smaller. Then OCR it.....but it seems all the characters have different structures.
That is mainly because the width/height of a 1080p screen is not an integer multiple of a NTSC/PAL screen. E.g. 1080/480 = 2.25, so one pixel in the NTSC equals 2.25 pixels in 1080p, so the same character is usually sampled differently at two different locations.
Then again, also the neighboring pixels influence each pixel when scaling, so even when scaling down by a factor of 2 or three, an "e" next to an "i" will be look slightly different compared to an "e" next to a "w" after scaling.

Anyway, this seems to an issue of SubRip with hires Vobsubs, so if SupRip is not an option, you should try to convince the author of SubRip to either allow 1080 resolutions for VOB/SUB or to implement import of BDN XML format, which is essentially only a PNG bitmap for each caption and an XML file with all the timing/additional info. Compared to parsing a typical binary transport stream, this is a piece of cake, so maybe he is willing to implement this.

Besides, I'm a little puzzled that none of the people here at Doom9 came up with the idea to tinker a script/Delphi Tool/whatever that OCRs the PNGs from an BDN XML export with a professional OCR tool and then puts together an SRT from the OCR output and the info inside the Xml.
0xdeadbeef is offline   Reply With Quote