Efficient indexing and similarity search using the Geometric Near-neighbor Access Tree (GNAT) for Face-Images Data

Abstract Efficient access to large multimedia data sets is urgent need for face recognition systems. The feature vectors that characterize face in images data are often high-dimensional. Due to the high dimensions of feature vectors and the large quantity of images in the data set, efficient and effective indexing structures are necessary to speed up searching and querying operation. In this paper we present an efficient face recognition system by combination features indexed into a metric structure GNAT. Its principle, consists to describe each face of database by A threes features Local binary pattern, Fourier descriptor and Histogram of Oriented Gradient, afterward we index a normalized combined features into a GNAT. We use a similarity measure calculated as an affine linear combination of L1 and L2 metrics. Our experimental results show that our technique is both correct and efficient.