Bisecting k-means clustering based face recognition using block-based bag of words model

Face recognition has received significant attention due to wide range of applications in information security, commercial and law surveillance. Bag of words (BoW) model, which was originally proposed to classify and identify documents, was recently introduced to the field of image processing and recognition and had got promising results. The authors (Li et al.) [1] who first applied the BoW model to face recognition proposed a block-based bag of words model for better recognition rate. However, as the number of the training face images increased, the block-based bag of words model took a long time to generate codebooks. In this paper, we propose a revised block-based bag of words model using bisecting k-means clustering method which could significantly accelerate the process of codebook generation. The revised model not only has very good recognition performance on face images from AR, ORL and FERET databases but also reduces the execution time significantly. We also applied the revised model to face recognition of the single sample. The experimental results show that the revised model is robust for illumination, expression, posture, partly occlusions, etc.