Choosing different sample intervals or times becomes one method of smoothing. So why would you think that highs and lows offer less noise? After all, they're only the highs and lows of fat-tailed, noisy data. Try an experiment: try your highs and lows charting based on different start times and see how they compare to each other. I believe you'll find it's similarly erratic to your original method.
A Bullish hammer with a long tail on the 5 min for example could look like 1 red bar followed by 1 greenbar on the 2 min . It's understanding the price action and context that matters. But yes your millage may vary with indicators and different time frames if you are using a bar close sensitive approach . Range bars might help fix the issue.
The likes of Steve Nison teach how to blend candles. When you have looked at them for long enough you start to see the patterns regardless of timeframe. Multiple brokers (demo is fine) is one option for different daily start times etc.