Open Neobatchief opened 9 years ago
Equalizing the histogram leads to a significant improvement in face detection in some situations, to what extent does it affect face recognition? I'd tend to think that face recognition would benefit from it as well since all the faces detected will be more similar in term of contrasts I think. Can you run some tests for the detection? Worst case would be to try and equalize the faces once extracted as well as the database faces to ensure a linear contrast across all of them (not sure how much the performance would be affected though).
Should we equalize the histogram when detecting the faces ? After that, the recognition is may be less effective