CSLBP and OCLBP local descriptors for speaker identification from video sequences

The aim of our work is to develop a speaker identification system based on the visual modality, in particularly the lip image, using local descriptors LBP and their extensions Center-Symmetric Local Binary Patterns (CS-LBP) and the opponent color LBP (OCLBP). An artificial neural network RBF "Radial basis function" is used to classify the correct speaker depending on what he is saying. Two databases are used for this purpose; the first one is our own audiovisual dataset where the recordings were made in speech communication and signal processing laboratory (LCPTS) concerning the Arabic monosyllables, the second one is the audiovisual dataset of Tulips concerning the English digits. A comparison for each descriptor and for each dataset is done to find the best solution for our lip characterization. Good recognition rate were obtained for CSLBP and OCLBP applied for the Arabic dataset.