I have been using FFmpeg to find the middle frame of a h264 video file, and extract the jpg thumbnail for use on a streaming portal. This is done automatically for each uploaded video.

Sometimes the frame happens to be a black frame or just semantically bad i.e. a background or blurry shot which doesn't relate well to the video content.

I wonder if I can use openCV or some other method/library to programmatically find better thumbnails through facial recognition or frame analysis.

Edit
Report