9/28/2023 0 Comments Subtitle edit ocr not workingThe RAM occupancy increases slightly from version to version. So that in version 10 it takes up to 2 GB of RAM during the first loading. But it gets worse from version to version, With the same operations, the results are similar everywhere. I checked previous stable versions 16, 15, 14 all the way to 10. I prepared the description for version 3.5.16 beta 222. Until now, this has worked without having to select this my opinion, Subtitle Edit is not properly managing the computer's RAM. In beta 184, this property is lost, and despite selecting a dictionary in the field, all text from the first to the last line is white background as if the dictionary was not specified (None).īackground recoloring is only restored when "Fix OCR errors" is checked. This property allows you to quickly locate the error line and its type visually. Lines with whole words that are unrecognized have a yellow background, while lines with unrecognized single characters have a brown background. In the picture above, with the selected language, the lines detected by OCR without errors completely have a green background for the text, I checked the version in Subtitle Edit beta 184 - the "All fixes" list is no longer populated for the case described above.Īnother problem arose - it concerns the "Subtitle text" window. That all corrections were made flawlessly. is empty and contains unrecognized words.įor those who do not know Polish, the good news is that The remaining tabs: and are filled in as expected. I also checked the stable version 3.5.15 - the problem does not occur. I checked other texts with and without italics - the problem is with all files. ![]() Regardless of whether I use pol_OCRFixReplaceList.xml or not. See lines 521 and 524 and it always happens The situation described occurs only for italic. "OCR auto correction" does not apply to the options you set.Īs you can see in the picture - except for the dictionary - the other "OCR auto correction" options are disabled,Īnd yet the OCR program made 13 corrections, although it should not. There are few guides on how to do it, for example this one.In the stable version of Subtitle Edit 3.5.16 and above. Even one frame per second will be more than enough so you can skip a lot of data by using frameskip (over 90% less data)Īfter collecting text entries you only need to remove repeating ones, leaving some margin of error as OCR is never perfect ("ą" can be detected as "a", "O" as "0" etc). Subtitles stay on screen for some time, you can skip a lot of frames.Subtitles are big, single color with sharp edges, with character border/shadow and sometimes solid background (unless video is of bad quality) - you can safely lower resolution of pictures by half or so. ![]() Only of part containing text (lower ~20%?). Subtitles stay (usually) in same spot on screen, so take "screenshot".It is possible to greatly shorten amount of time needed to process video by limiting size and amount of pictures (data). (Or you can use OpenCV and create your own soft, it is quite easy for such complex task) You can overcome it by feeding it frames from video file. However most of OCR software you can find for free is able to process only images. Hardsubbed text is merged into video stream, so OCR'ing is only way to extract it. What you are looking for is OCR software.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |