Weighted Majority Voting for Face Recognition from Low Resolution Video Sequences

Abstract

In this paper a new system for recognizing faces from video sequences using weighted majority voting (WMV) method is proposed. In the training phase, the system uses principle component analysis (PCA) based single eigenspace generated by sequences of faces of all subjects with the same resolution. For the testing phase, the system employs several preprocessing tasks whereby for all subjects' videos, the face images with varying resolutions in different frames are automatically extracted, histogram equalized to alleviate the effects of changing illumination, and upsampled to the resolution of the eigenfaces. For the recognition phase, each recognized subject is assigned a weight based on a measure of information capacity of each tested frame. Finally the subject with highest cumulative weight, through the video sequence is declared to be the recognized person. The proposed WMV system is robust to scale changes and effectively addresses the problem of recognition from low resolution video sequences.