Add memorization to the bitMatrix class & return the transformation

[LHDi]

Increased performance because we don't recreate the matrix for every scene.
the performance is way better for continuous streaming.
return the transformation matrix.

[danimoh]

Note that the avoidance of buffer memory allocation and garbage collection for each scanned image is also achieved by #115.
#115 has the additonal advantage of no memory overhead, as no additonal memory has to be allocated or cached, in contrast to the Memo cache here that is on top never pruned.