* without border extrapolation
* with aligned write
* process 4 pixels per thread in 8u case

Signed-off-by: yao <bitwangyaoyao@gmail.com>

Issue #2231: Attempt to fix segfault, when copying labels. LDA needed to be updated to generate the Wrapper correctly.

add channel 3 support
add fast way Between CPU and GPU for the data which is aligned

Signed-off-by: yao <bitwangyaoyao@gmail.com>

facerec.cpp: Removed Original LBP wrapper, because it is not in use and therefore generates a warning.

facerec.cpp: Added an exception to the LBP extraction, so now a meaningful exception is thrown if the wrong image type was given. Thanks to Eric Christiansen for reporting.