Method and system for determining ethnicity category of facial images based on multi-level primary and auxiliary classifiers

A method for determining ethnicity class based on facial images, comprising the following steps of:

• a) setting up multi-classifier architecture,
• wherein each classifier in the multi-classifier architecture, called an auxiliary class machine, is tuned to a specific auxiliary class and is specialized to classify ethnicity within the specific auxiliary class,
• b) training auxiliary class machines so that a first output of an auxiliary class machine is a likelihood of a given facial image belonging to an auxiliary class and a second output of the auxiliary class machine is a set of likelihoods of the given facial image belonging to respective ethnicity classes,
• c) extracting image features of faces detected from an input image frame in a way that the faces have standard position, size, and orientation,
• d) processing the image features using the multi-classifier architecture utilizing information about demographic classes of the faces other than ethnicity, called auxiliary classes, to determine the ethnicity classes of the faces,
• e) aggregating first outputs and second outputs from the auxiliary class machines for the faces,
• f) computing ethnicity scores of ethnicity classes by weighting ethnicity outputs from the auxiliary class machines by auxiliary class likelihood outputs from the auxiliary class machines,
• wherein the first output is auxiliary class likelihood output and the second output is ethnicity output, and
• (g) applying a voting scheme to a series of multiple classification instances, producing a series of final ethnicity vectors for multiple input facial images of a person,
• wherein each ethnicity label that has the highest score from each of the series of final ethnicity vectors gets one vote that is contributed toward a final decision for determining ethnicity class.