Multi-Classifiers Face Recognition System Using LBPP Face Representation

Abdellatif Dahmouni, Nabil Aharrane, Karim El Moutaouakil, Khalid Satori
2017 International Journal of Innovative Computing, Information and Control  
Local Binary Probabilistic Pattern (LBPP) is a local descriptor able to improve the recognition capabilities of a typical pattern recognition system. It is a new alternative of the famous Local Binary Pattern (LBP) descriptor based on confidence interval concept. To achieve an enhanced representation for face's principal components, LBPP evaluates each current pixel using a probabilistic confidence interval related to its neighborhood. In this paper, to improve face recognition performance we
more » ... opose a new methodology based on the combinative use of LBPP descriptor, Two Dimensional Discrete Cosine Transform (2DDCT) frequency subspace, and some machine learning algorithms. The main idea behind this methodology is to elevate the weak points of each one of them, while making use of their major advantages. Hence, after the LBPP processing phase, 2DDCT method decomposes obtained image into set of local features vectors. Each local vector is formed by the k-first zigzag coefficients for each sub-image. Then, we carefully concatenate all local vectors into a single features vector. In addition, obtained features dataset will be classified using relevant machine learning classifiers. To access our solution, we applied it on ORL, Yale and AR face databases. Obtained results clearly show the effectiveness of the proposed approach compared to the existing state of the art techniques. Indeed, the LBPP capacity to discriminate face components, the small size of 2DDCT features vector, and the efficiency of used classifiers, allow justifying the proposed approach's good performance.
doi:10.24507/ijicic.13.05.1721 fatcat:rfc3asx3t5cqhmfj4keuckyrxi