psyche

I think we’ve all seen the original seam carving video, but I hadn’t seen the following. It deals with problems in the original algorithm stemming from removing low-energy seams causing there to be more edges in the retargeted image - this roughly means that you can potentially make an image more jagged despite only altering the least edgy parts. Their solution was to consider “forward energy” and remove seams which, when removed, would insert the least amount of energy/edginess/contrast, which means generally the retargeted images are now even better quality.

They also treated 2D video as 3D objects, with the third dimension as time. Have a look:

I still find their work strikingly elegant. Not only is it effective, with many potential applications, but it’s also clear enough for us to get the gist of without reading their paper or doing any deep thought. It’s fantastic! If it’s proliferated, it’ll be even harder to tell what’s real and what’s not (e.g. is that field really that big, or are those inserted seams? Is that street really that narrow, or…?) unless it’s carefully managed. Right now we only have to contend with Photoshop (I say only but that’s only because people don’t abuse it as much as it could be abused, or with as much skill [generally] as possible)…

Pax