I think that it searches for the "nearest" power-of-2 (1/2, 1/4th, 1/8th...) subsample of the original bitmap which "better fits" the given maxWidth and maxHeight.
3543/293 = 2215/277 = 8 aprox (this is the subsampling)
Also, the scale factor of the subsampled bitmap remains (3543/2215 = 442/277 = 1.6 aprox )
As in your case you gave square dimensions for a non-square original bitmap, it made its best
