CUDA Accelerated Face Recognition Using Local Binary Patterns
Istanbul Technical University, Department of Computer Engineering, 34469, Turkey
Istanbul Technical University, 2012
@article{tek2012cuda,
title={CUDA ACCELERATED FACE RECOGNITION USING LOCAL BINARY PATTERNS},
author={Tek, S.C. and G{"o}kmen, M.},
year={2012}
}
In this paper, we present a GPU accelerated face recognition framework using CUDA. We use weighted regional LBP histograms as features and k-nearest neighbour (k-NN) algorithm for classification. Our first contribution is to present an efficient way to compute LBP values from an input image and construct weighted regional LBP histograms in GPU using a single kernel. The second contribution we make is to propose a massively parallel GPU implementation of the k-NN algorithm optimized for handling high-dimensional feature vectors. Comparisons with CPU implementations have shown that, by accelerating both the feature extraction and classification process of the face recognition algorithm, we have managed to achieve up to 29x increase in recognition speed.
March 10, 2012 by hgpu
