For 3d, collision is best done with z-buffer... which could be done with your map but would require a corresponding z-texture for checking.

Closest with 2D is a separate collision map for objects, for the 2 reasons of size and the fact that textures are stored in texel format (which you'd have to decode every check).

Unless you have another reason to have more detail, you could probably create a 1bit resolution collision map of hit/miss (1/0), or scale it up to whatever degree you find easiest. More memory/less work would be a black/white map that is just stripping the alpha channel of your image and storing in in the lowest comfortable color resolution.

GX has functions to check Z and RGBA on the buffer, but those are total of everything written, not just your texture. Collision isn't really a graphics thing anyway, since based on design pattern it falls in the model/logic, not viewing category.