Design And Implementation Of Classification Algorithm Based On Low Rank Factorization

Classifier design is one of the most important problems in face recognition.Extracting effective feature representation from high-dimensional face image data sets is the key to improve classifier performance.The low rank decomposition of matrices has become the most effective and widely used method for extracting low dimensional feature representation.Through low rank decomposition algorithm,we can find the correlation between face images,and recover the intrinsic low dimensional structure of face image dataset.In this paper,the algorithm and implementation of classifier based on low rank decomposition are studied in detail.1?Laplacian Regularized Non-negative Sparse Low-rank Recognition ClassifierIn the real world,data often reside on low-dimensional manifolds embedded in a high-dimensional ambient space.The existing non-negative sparse low-rank representation classification fails to consider the nonlinear geometric structure between the training samples,which makes data lose the information of locality and similarity in the learning process.In view of the above problems,this chapter proposes a non-negative sparse hyper-Laplacian regularized LRR model(NSHLRRC)which can not only reveal the global low dimension structure,but also capture the intrinsic nonlinear geometric information of the data.Experimental results on AR face database and CMU PIE database verify that the proposed method has good classification performance.2?Weighted Low-rank Representation Classifier for Face RecognitionThe traditional low rank representation algorithm usually uses the way of solving the standard kernel norm to get the rank of low rank matrix.The standard kernel norm is the sum of the singular values of the calculation matrix,but the rank of the matrix is the number of non-zero singular values,so the traditional low rank representation algorithm ignores the relationship between the samples in the data set.In order to solve these problems,this chapter propose a weighted low-rank representation classifier(WLRRC)for face recognition.The low rank matrix is more robust by assigning different weights to different singular values.Experimental results on ORL face database and Extended Yale B database verify the performance of the proposed algorithm is better than the existing low-rank representation classifier algorithm.3?Weighted Non-negative Sparse Low-rank Representation ClassificationBecause the sparse constraint can improve the identification ability of the classification algorithm,the non-negative constraints can increase the interpretability of the algorithm.In this chapter,a weighted non-negative sparse representation classification method(WNSLRRC)is proposed.With the weighted low-rank representation classifier compared,the algorithm has stronger discrimination and interpretation.Experimental results on ORL face database and AR face database show that this method has good classification performance.4?Design and implementation of face recognitionThis system is based on the python platform.It uses a variety of techniques such as ipywidgets,opencv,numpy,matplotlib,and dlib to realize face detection and recognition for different races in a complex background.System functions include:(1)Finding and selecting pictures in the server through the drop-down box;(2)Face detection,including single-person detection and multi-person detection;(3)Face recognition:feature extraction on the face,and then based on the picture library The training samples in the comparison are compared and the classification results in a recognition result.(4)Real-time face recognition:calls the camera to the target for face feature extraction,and then bases on the input image and achieves the classification recognition results.