Automatic photo tagger

Abstract

Looking at the case of present social networking websites and photo sharing, we present the problems faced when implementing a face recognition system in such an environment. One of the main problems is the huge size of the database, possibly with millions of registered users. We see that this can significantly degrade the performance of a face recognition algorithm. Therefore we propose some ways in which the performance can be improved when considering a social network environment. We do not use a large database for testing, but instead measure the factor difference of the processing speed when applying proposed techniques. Ratios or measurements of physical features and the determining of a face's gender can both be used to categorize the data set and only use a subset when classifying. These reduce the number of comparisons required, improving the speed of the algorithm, and at the same time improve or degrade the algorithm's accuracy. The system implements two face recognition algorithms for comparison, along with the gender recognition and facial features measuring