Research On Face Recognition Based On Bag Of Words Model

With the rapid development of computer and network technology, face recognition technology has gradually become one of the hot research areas in the image analysis and understanding and pattern recognition field. Face recognition has been widely used in such aspects as financial security, the national information security and public security. Although the face image is easy to collect, due to the existence of the problems such as age, the expression, light, shade, posture, etc., the progress of facial recognition technology is slow,the most present face recognition methods try to isolate to solve only one difficult problem in face recognition without considering the solution of the case that they appear at the same time,so a robust face recognition methods will be the research key content in the future. Bag of words (BoW) model originally applied in the field of document processing, and then has been introduced into the field of image processing by the researchers in the field of computer vision, and got good performance. Bag of Words model was applied in face recognition area in this paper. Based on the study of the technical, improvement has been done in order to overcome the disadvantages of the model.First, in order to make a more reliable, more efficient visual dictionary, based on the study of traditional k-means clustering method, the bisecting k-means (BKM) algorithm is applied to BoW model in order to overcome the defects of the traditional k -means clustering method, such as the clustering performance strongly depends on the initial clustering centers,easy to encounter local minimum problem and not suitable for large databases. However, in the process of each iteration, the BKM method only renews two cluster centers. Comparing to the basic k-means algorithm, it can avoid the uncertainty in selecting the initial centers, reduce the runtime significantly. After the iteration, the BKM also can avoid falling into a local optimization result. The experimental results show that the proposed method not only improves the face recognition rate, but also significantly reduced the codebook generation time.Second, this paper applied the improved method to face recognition of the single sample.In real life, a lot of the times we can only obtain one single image of the samples, such as identity card, driving license, passport, etc., so the single sample of face recognition system is a developing direction of the future. The experimental results shows that the proposed algorithm for single sample face recognition rate is very high, and the proposed algorithm is robust for illumination, expression, posture, shelter, etc.Finally , based on the analysis of the support vector machine (SVM) classification method and face recognition application requirements,relevance vector machine was used to classify face images. The proposed method overcomes the problem of the predicted results are not the probability characteristics and it must satisfy the kernel functions, which are existed in the support vector machine. Compared with support vector machine (SVM), relevance vector machine is more sparse, using the fewer number of support vectors, thereby significantly reducing the computation time of the test samples. The experimental results show that the recognition rate of relevance vector machine is slightly lower than support vector machine(SVM), but on the testing time, relevance vector machine spend a little time, almost achieved the purpose of real-time testing, which has great significance for the practical application of face recognition.