## Formalize statefilter estimation relationships

For a frame I(y,x,z_hat,w_hat;t), if we want to estimate the corresponding tensors **I**(y,x,z,w_hat;t) or **I**(y,x,z_hat,w_hat;t), the input frame needs to have a correspondence to the output frame. The best way would probably be to attach all relevant coordinates to the image frame. To explain, the difference between I1 = I(y,x,z_hat) and I2 = I(y,x) is that, even though both can be considered as having an infinite number of additional singleton dimensions, I1 has 3 COORDINATE dimensions (Y,X,Z) and I2 has 2 (Y,X).

I2 can be made into I1 by 'naming' an additional dimension.

If we're estimating a tensor **I**(y,x,w) from a series I(y,x,w_hat), then the sliding window approach would effectively do find(w==w_hat) and update the tensor estimate correspondingly.

All this said, the MSOT metadata gives us the 'natural' coordinates of the entire sample space in the form of meta:ZPositions and meta:Wavelengths.