FREC- a flexible Face RECognition system
FREC is written with C++ and OpenCV on Linux. It was written for the purpose of testing with different face recognition algorithms. Image pre-processors also can be easily plugged in. Currently it features a couple of basic pre-processors eg. histogram equalizer , image detector etc. and EigenMethod based recognition system.
FREC loads the database of images according to its configuration files and passes them to the pre-processing and recognition systems. Depending on the run mode, the recognition system can train its internal structures and save the required data to be used during normal operation. During normal operation, recognition system can load this data and perform recognition operation.
FREC also gives an interface to write a self test of the recognition system. For example, EigenMethod based recognizer provides comparison graphs in the form of gnuplot scripts normalized Eigen Faces as output during self test.
For testing I have used the face database at [] which consists of [] images and around 20 images per person.
EigenMethod based recognition system:
During the self test mode, the recognition system will train itself with all images excluding one image at a time and compute the accuracy by taking that image as input. This is carried out for each image per person for each person in the database.