algorithm - Using Gaussian Window to subdue the background in the facial recognition process -

i trying perform facial recognition set of 28 database images (4 per person 7 person). each person has 2 images each taken in 2 different backgrounds. every image has face of individual @ center of image.

initially when ran facial recognition algorithm images of 7 people (2 each, in same background) algorithm worked fine , dandy. however, upon addition of face-images in different background, algorithm did not work intended, signalling fact background in images causing problems.

i read paper matthew turk , alex pentland , suggest multiplying images 2 dimensional gaussian window centered @ face. tried doing so, however, performance of facial-recognition algorithm did not improve.

could possibly me understand process associated two-dimensional gaussian window , how applies facial recognition? turk , pentland claim process worked them.

i used following formula 2 dimensional gaussian pdf @ pixel location (x,y):

it matthew turk , alex pentland talking filtering image 2d gaussian matrix, though not familiar paper. filter can thought of either convolution in time domain -or- multiplication in frequency domain. if filter image 2d gaussian, effect smoother version of image. see below:

consider original (unfiltered) picture:

filtered gaussian matrix:

with result 'smoothed' version of original:

let me know if helps!