Non-negative matrix decomposition (NMF) algorithm is the decomposition of all the elements in the matrix under the condition that each element should be non-negative. As a relatively effective technique for dimensionality reduction, NMF has been widely applied in the area of mathematics-physics, engineering and image feature analysis. However, there are few systematic reviews on NMF, especially the application of NMF in image feature extraction. In this paper, NMF algorithms are classified into standard NMF algorithms and improved algorithms according to the theory and application characteristics of different approaches. The basic principles, advantages and shortcomings of these NMF algorithms are systematically analyzed and compared. Firstly, the basic idea of non-negative matrix factorization is introduced, and its application in image feature extraction is illustrated by taking face image as an example. Then, the basic methods and improved algorithms of NMF are emphatically discussed in detail. The examples of local features extracted by different NMF methods are demonstrated on the basis of object feature analysis methods. Finally, the problems to be solved in the practical application of NMF are put forward for improving.