Research On Face Recognition Algorithm Based On LBP Feature Extraction

The technology of face recognition is a kind of identity recognition method that uses the computer to analyze the face images and extract the effective feature to discriminate identity.Because of its great application prospect,it has become a research hot spot in the pattern recognition field.In this technology,how to extract and select effective features to descript the face image is a research focus.As an effective algorithm of feature extraction,LBP has showed great performance in face recognition,but it still exist some deficiencies.So this paper studies the LBP algorithms,and then introduces improved methods as follows:The definition of the direction of the D-LBP operator based on LTP operator,D-LTP operator is proposed to describe,in reducing the LTP feature description operator dimension at the same time using the LTP value of three D-LBP encoding to solve the same direction and the reverse is unable to distinguish between the same side of the problem by using the D-LBP operator.At the same time,because of the characteristic dimension of D-LBP,the global gray feature coding is introduced to make the algorithm have the ability to describe the global features.The SD-LTP algorithm is proposed by introducing the adaptive threshold t_L and t_G,which makes up the local feature extraction caused by the fixed threshold,and also enhances the anti-noise interference ability of the algorithm.Experiments on ORL and FERET show that SD-LTP effectively solves the problem of LTP algorithm in data dimension and threshold selection,which greatly enhances the classification ability of the algorithm and confirms the improved validity and feasibility.The Gaussian white noise experiment on the ORL library also demonstrates the anti-noise of SD-LTP.In order to solve the limitation of LBP algorithm for multi-scale feature extraction,the SD-LTP pyramid feature is introduced to solve the limitation of LBP algorithm for face multi-scale feature extraction.Experiments on the ORL and FERET libraries show that the SD-LTP pyramid feature is superior to the traditional algorithm and higher than the original SD-LTP algorithm,compared to the superiority of the classical face recognition algorithm in recognition rate.The experimental results show that the SD-LTP pyramid feature still has a good classification ability in the case of a small number of training samples,which embodies the SD-LTP pyramid feature for a variety of training examples.Noise robustness.