Vijay's profile - activity

ZachGarner, thanks for your comment. Ya, I have looked into the tutorial. Author was just guessing that it could be FERET. He did not confirm exactly on which database OpenCV face detector was trained on.

Face verification and recognition are two different problems. Recognition is a multi-class problem where the goal is to assign the test image a known label from the database while verification is a binary problem where you need to say whether the given two images are similar or not.

OpenCV fisher faces is aimed at recognition and always assigns the test image a label from the database. However, it may not perform well on verifying unknown faces since the fisher face algorithm needs the training samples of all the subjects.

Face verification is getting lot of attention these days. Many algorithms are introduced in the last few years and tested on challenging LFW dataset. I would advice you to take a look at implementations available at http://vis-www.cs.umass.edu/lfw/results.html

Could anyone please reply to this question?

I want to compare the performance of HoG-SVM (dalal-triggs) detector and Viola-Jones on faces. In order to have fair comparison, I would like to use the same training data for training HoG-SVM detector that was used for OpenCV Viola-Jones cascade. I tried to look into the openCV documentation and could not find the details of the training set used for generating haar_cascade_frontalface.xml ? I also looked into the paper [1] but it only says that around 5000 positive images have been used for training but does not specify the name of the dataset. Could anyone provide the details of the dataset used for Opencv frontal face cascades?

[1] Rainer Lienhart, Alexander Kuranov, Vadim Pisarevsky, Empirical Analysis of Detection Cascades of Boosted Classifiers for Rapid Object Detection, Tech Report, MERL 2002.