An Improved Confidence-Based Boosting Face Recognition Algorithm under Large Pose Variations

one of the significant remaining challenges in face recognition, which has attracted much attention, is face recognition under various poses. So, this article introduces a new method for face recognition under various poses. The proposed method attempts to address the problem by introducing a new Confidence-based boosting algorithm to improve the performance of the tied factor analysis (TFA) method called confidence- based tied factor analysis (CTFA). In the present work, the confAdaboost.M1 algorithm is applied on the TFA generative method, which obtained state-of-the-art face recognition performances on large pose variations. Actually, the TFA is regarded as a base classifier or weak learner in the ConfAdaboost.M1 algorithm. The training data likelihood weight is updated by the ConfAdaboost.M1 algorithm. In the proposed method similar to the TFA, at the recognition step, a face image is used at a non-frontal pose. Then to improve the performance, the Gabor filter is applied at the preprocess step. The new method has evaluated on the FERET database and compared with the original TFA method in recent studies, the results of which have demonstrated superior performance under large pose variations.