Fusing Local Binary Patterns With Wavelet Features For Ethnicity Identification

Abstract

Ethnicity identification of face images is of interest in<br> many areas of application, but existing methods are few and limited.<br> This paper presents a fusion scheme that uses block-based uniform<br> local binary patterns and Haar wavelet transform to combine local<br> and global features. In particular, the LL subband coefficients of the<br> whole face are fused with the histograms of uniform local binary<br> patterns from block partitions of the face. We applied the principal<br> component analysis on the fused features and managed to reduce the<br> dimensionality of the feature space from 536 down to around 15<br> without sacrificing too much accuracy. We have conducted a number<br> of preliminary experiments using a collection of 746 subject face<br> images. The test results show good accuracy and demonstrate the<br> potential of fusing global and local features. The fusion approach is<br> robust, making it easy to further improve the identification at both<br> feature and score levels.