Finding and Using Moving Images In Context

Yesterday’s question for Gary Geisler, PI on Open Video Digital Library Toolkit (OVDLT), an IMLS grant in which NHF participates. Figured Gary, at UT Austin, would be up to date on the literature on best practices for identification of video through single images, thumbnails.

What is current thinking on thumbnails from video?
How to weigh selection of first frame of picture versus a “content interpretive” frame (necessarily subjective)?
What are you thinking for thumbnail size and identifying text?
Will you boost contrast or do other manipulations?

–mostly b&w in our OVDLT sample.

And all b&w in the Branch and Gilbert selections.

An example of thumbnails in WGBH’s OpenVault. Small, color, and supported with much text.

______________________________________

Update, July 11, Gary Geisler writes:

I’m fairly familiar with the research literature in the area of digital video and I’ve never seen any discussion of thumbnail size or the details of which frames to select. There might be a study or two out there on these topics but I don’t think the area has really progressed to the point where people have looked too closely at these sort of details (though they would be interesting things to look at).

For the Toolkit I’ve been working on a keyframe extraction tool, based on code from CMU’s Informedia Project, that selects frames based on scene changes. My plan is to enable the organization using the tool to select their own size dimensions, though we will probably suggest a default value. For frames used in storyboards I think a width of 100 or 110 pixels works well (the size we used on open-video.org is 86 wide, but that is a bit too small in my opinion). I don’t have any plans to boost contrast or do other image manipulations. An organization could certainly do that to the keyframes that our extraction tool has produced, but trying to build that into a tool is not feasible, I don’t think, since each video could have different characteristics requiring different manipulation parameters and you really need human intervention to determine what would be optimal for each video. This might be a good thing to mention in the documentation — just let people know that they could run their keyframes through Photoshop or something if they want to boost contrast or something after the frames have been extracted by the tool.