Face recognition using LBP, nLBP and αLBPalgorithms

Face recognition is of great interest because it is one of the most important image processing applications. Although the success rates of the studies in the literature are high, the performance in out-of-control situations is still not better than human. There are many challenges to design an accurate and robust face recognition system, especially in non-restricted environments. In this thesis classical local binary pattern (LBP), neighborliness local binary pattern (nLBP) and αLBP were used for a face recognition problem. nLBP is formed according to the relationship between the neighbors around each pixel. nLBP has a distance parameter which specifies the distance between consecutive neighbors to be compared. Different patterns are obtained for different d parameter values. αLBP operator calculates the value of each pixel according to an angle value. Angle values can be α = 0, 45, 90 and 135 degrees. The ORL face database was used to test the proposed approaches. nLBP, αLBP and classical LBP features were extracted from face images and classified using k-nearest neighbor (kNN) and artificial neural network (ANN). 98.25% recognition rate was obtained using kNN with nLBP. A recognition rate of 88.50% was obtained with αLBP using ANN. The recognition rate of 83.50% was obtained with classical LBP. The proposed nLBP and αLBP approaches were found to be more successful than the classical LBP method. In the literature, the success rates obtained in the studies performed on the same ORL face database were compared with the success rates of the proposed approaches in this thesis study. As a result, the proposed LBP-based approaches achieved significant success in face recognition.