Noise Tolerant Histogram Voting for Gender Classification Based on LBP

In this paper, we present a noise tolerant descriptor based on a local binary pattern (LBP) method. Due to threshold-based operations, these types of LBP methods are sensitive to noise factors. The use of a robust LBP (RLBP) reduced some noise effects. However, it may lead to a loss of subtle local texture information. Instead of concatenating the LBP and RLBP features, we produced a histogram as a weighted sum of the histograms of the LBPs and the RLBP. The proposed noise tolerant LBP (NTLBP) was calculated using the LBP histogram and histogram voting results of the RLBP. Without increasing the number of features, NTLBP proved to be robust against noise effects. We conducted several gender classification experiments using the FERET database and the NTLBP outperformed both the LBP and the RLBP methods.