Investigating a problem with 3D Registration using the Mutual Information measure(s):

During the Finetune stage (which is 100% multi-threaded), Intel Performance monitor indicates only about 30% load of 8 threads (i.e. 2.4 threads on average).





Update: Was able to substantially speed up the Finetune stage for the Mutual Information calculation.  Correspondingly the processor load is at 58% at this stage.

However, the principal part of the MI can not be parallelized efficiently due to the summation of the large Histogram memory object.

FireVoxel Build 125B is released.

1. Restored the Initial Transform functionality in 3D Registration. It combines correctly now with the 3D Polynomial transform.

2. Added the Automatic Build Test subtest for 3D Registration with the Initial Transform

3.  Version Expiration date is properly displayed in the About Dialog box.

Fixing registration defect reported by Dr. Rsinek:

FV124 crashes when running 3D coregistration in several configurations:

 Try to reproduce these steps:

1) open raw:    parkp_rt.qr.im
2) Layer control ; open raw :   parkp_rt.qr_targetROI.im

Similar result was obtained with "Mutual Information Normalized".   The "Noise" level that defines binning of the histogram for Mutual Information can vary in a wide range from [20,1600] with a good result.