I look at fractals or price structures I recognize from all the chart/screen time over the past 28 years. Or setups.
You can, you can create a sliding focal window of a certain length on a timeframe chart, the algo would take automatic snapshot as it slides along and than waits and takes a calculation of the result. Eventually only those price structures that resulted in significant variance would have a bias and utility. The algo could be programmed to trade.
I had trading buddy in in 1999 who had 100 or so transparencies of daily patterns (5 min chart). He'd lay them on top of the current day every hour till he had a match and would then enter a position. He always did pretty well trading into the close when he had 4 or 5 hours line up.