face_align.py not consider multiple face bbox

[gangeqian]

I can not get the same result on LFW 112x112 or 112x96 aligned dataset when I use face_align.py, about 96% or 97%。 Then I found the face_align.py only use the first face bbox when MTCNN found multiple face， which results in error face label in generated LFW dataset。

After processing multiple face (get the max bbox arear one), I get ACC: 99.2 on LFW 112x112 dataset using Mobilenetface