Research On Local Binary Pattern Face Recognition Algorithm By Using K-means For Dimension Reduction

As a typical pattern recognition problem,face recognition mainly involves image acquisition,analysis,processing and classification,etc.With the rapid development of science and technology,face recognition technology has been further developed,but some problems still remain to be solved.As a key part of face recognition,feature extraction largely affects the system performance directly.In recent years,the local binary pattern algorithm has been widely used because of its excellent texture description ability,and the method of extracting the face feature by using the local binary pattern has also become one of the popular researches.However,the local binary pattern algorithm has its limitations.In this paper,it is deeply studied and improved from the following aspects:(1)For the problem that the local binary pattern algorithm extracts the local features which are not complete and can not extract the global features,two algorithms,floating local binary pattern and high dimension local binary pattern,have been proposed to improve the algorithm.As the improvement of these two algorithms are different and have their own advantages,so that a exceed local binary pattern algorithm has been proposed by fusing the two algorithms above.The exceed local binary pattern algorithm not only has the ability of extracting neighborhood edge features of floating local binary pattern,but also has the ability of extracting global features of high-dimensional local binary pattern.Experiments show that the recognition rate of the improved algorithm is higher than that of the classical local binary pattern algorithm,and the recognition rate of exceed local binary pattern algorithm is the highest.(2)For the problem of multiplication of the dimension of feature vector in the three algorithm above,a method of dimensionality reduction based on K-means algorithm has been proposed.The clustering method is used to fuse the features with high similarity to reduce the dimension of the feature vector.In order to solve the problem that K-means algorithm is very sensitive to initial clustering central point,according to geometric distribution of face feature vectors,a uniform K-means algorithm has been proposed in this paper.Combined with the variation trend of the K-value and overall error in the iterative process,a new evaluation function has been defined.Using the evaluation function to select the number of initial points,and then determine the initial clustering center by a way of uniform selection.The experimental results show that the stability and overall error of cluster of the uniform K-means algorithm are better than classical K-means algorithm;The comparison experiment with the PCA algorithm for dimension reduction shows that the performance of dimension reduction of uniform K-means algorithm is better than PCA algorithm.In this paper,by studying the calculation method of the local binary pattern,a exceed local binary pattern algorithm based on uniform K-means for dimension reduction has been proposed.The face feature vector extracted by this algorithm is not only low in dimension and strong in capacity of characterization,but also improves the recognition rate of classical local binary pattern algorithm under the condition which guarantees the stability of the recognition time of the algorithm.