A mean filter is one precise type of linear filters that replaces a value by a weighted combination averaging the neighboring values. A constant offset value, or an affine trend, is "quite" invariant by an averaging filter. Therefore, it may estimates a constant or linear baseline well.
Trend removal might be less well-defined. For the mean, there are many alternatives, like the exponential weighted moving average (EWMA), which is an IIR filter.
Baseline is even less well-defined. It can dwell between simple instrumental offets and complicated "references" that may include noie.
The last one is the fuzziest to me. But there is a notion of oscillation, that could be incorporated into a filtering model with knowledge of the signal. Tracking and informed filtering (eg Kalman) can be useful for that purpose