This is not about DVD subtitles, which are images as you say. This is about “Line 21” closed captioning. I.E. the text data that is embedded in an analog tv signal. There should be no OCR needed.

We use DB2, but mostly because we call it from a mainframe.

Typically, the last mile is often not fiber. Telecoms are generally required to continue non-voip land line service. Its one of those regulations that, despite having good intentions, end up causing issues.