Edit: TLDR; Partly automated hair selection is complex and lacks truly robust methods using only the tools available in photoshop.
That is not a usable method. First the gamma encoding used in typical RGB profiles is problematic here, and you have no suitable way to get rid of that. Converting to a different color space won't do anything for you.
(This is a really rough explanation. It certainly contains some errors, at least errors of omission. )
The other problem is based on the way this is constructed. It's typical to use gaussian blur, so with respect to the color space, you generate something which is normally distributed when examining the image across any local minimum or maximum value. Local maxima and minima here are just pixels that are brighter or darker than any neighboring pixel. The high frequency layer is equivalent to taking the image and subtracting its normally distributed component. This implies to someone looking at the two images that significant residual deviations would indicate strong edges.
The strong edges thing holds to some degree, but photoshop's methods won't help you set the appropriate things to high and low. Even outside of photoshop this isn't considered to be an ideal method at all. If you tried to use the high frequency layer, it would require the use of propagation methods of edge finding, which aren't really available in photoshop. The other types would require significant modification. Even then it won't necessarily work well, because normal distributions are not an ideal way of describing image patterns.
For hair, use a brush. Tune the hardness. Paint in flyaways, make color corrections, and fill gaps as necessary. The forum gets a pretty wide range of users, but if you don't understand this, then trust me, it's a bad idea