Study On Image Recognition Method Based On Non-Negative Matrix Factorization

With the rapid development of computer science and Internet technology,a huge amount of image and video data is produced every day.How to organize,process,retrieve,and apply these image information has been the key research issue in the field of artificial intelligence.While image recognition,as the main research content of pattern recognition and image processing,has found significant applications in many areas.In view of the drawbacks of inaccuracy of the traditional image recognition algorithms,this paper studies the image feature extraction and recognition methods based on convolutional neural networks and negative matrix factorization network.The main work of this paper is as follows:1.The structure and characteristic of convolutional neural networks are studied and the image features are extracted via convolutional neural networks.The nonnegative matrix factorization algorithm and its extension in kernel space are summarized.2.An image recognition method based on semi-nonnegative matrix decomposition network is proposed.On the basis of principal component analysis(PCA)network,a new filter bank is constructed by using semi-nonnegative matrix factorization algorithm instead of PCA.In addition,we use the 7)norm regularizer to restrict base vectors to the substitute the nonnegative constraint on basis matrix.Compared with the principal component analysis network,this network improves the recognition performance in three face databases and standard handwritten database.3.An image recognition method based on kernel semi-nonnegative matrix factorization network is proposed.The semi-nonnegative matrix factorization is mapped to kernel space on the basis of semi-nonnegative matrix factorization network.And further obtain the nonlinear information between data.Compared with the principal component analysis network,this network improves the recognition performance in three face databases and standard handwritten database.