Created a ridged registration function that returns a normalized covariance matrix back.  Seems to be off by a factor of 0.5.

Gaussian filter no longer has quantization error between dimensions.  Type casting happens now at the begging and then at the end instead of in the kernel.

Cuda sum does not send data to the device if the number of elements to add is less then a block size

*********Fix convert type*********

*****Convert Type Yet Again******

******Commit to fix another Commit ***********