Face Recognition Based on Multi-scale and Double-Layer MB-LBP Feature Fusion

Aiming at the reason that traditional face recognition algorithms are greatly affected by factors such as illumination and noise, a feature extraction method based on multi-scale and double-layer MB-LBP operator is proposed. First, extract three MB-LBP features with scales of 1*1, 2*2, and 3*3, respectively. Based on three different scale MB-LBP feature pictures, the MB-LBP features of the second layer are extracted, and then the respective histogram features are counted and all features are combined into a higher dimensional feature. Principal component analysis (PCA) is used to reduce the dimensionality of the fused feature data. Finally, support vector machine (SVM) is used for classification. By testing on ORL and AR face databases respectively, this method has significantly improved the accuracy of face recognition compared to traditional methods, reaching 99.5% and 99.2% respectively. In addition, in order to verify the adaptability of the algorithm to light and noise, through the brightness adjustment and noise addition to the face data set, and then test and verify. while the recognition accuracy of the traditional method has dropped significantly, the recognition accuracy of the algorithm is still more than 98%.