I have used tesseract ocr combined with imagemagick and ffmpeg to great success for video text extraction.
Can you list your script/pipeline? I haven't had much success (though, I only ran ffmpeg's internal tesseract OCR[0], no imagemagick processing or any other processing in between)