A Non-negative Matrix Factorization Method Research And Application In Blind Source Separation

At present, the signal processing, data analysis and data mining is widely used in scienceand engineering field. From the original data set, observation and understanding of the complexdata and extracting the knowledge have become interested in present important challenges andgoals. Decomposition of matrix in signal processing and linear algebra fields belong to a veryimportant issue, among them the negative decomposition of matrix is D.D.Lee and H.S.Seungput forward in1999with non-negative constraints of the matrix decomposition method. Data thenegative meet the authenticity of the physical signals, while the negative will lead to its have thecorresponding sparse solution. The sparse data to a certain extent in description will suppressexternal environment interference, such as illumination change, keep out or object of imagerotation, etc. Therefore, the gradually become a non-negative constraints based on with sparseconstraint of nonlinear dimension about reduction method.This paper first introduced the nonnegative matrices and the improved algorithmdecomposition, however in image blind source problem, had the nonnegative matrices and theimproved algorithm of image decomposition blind source separation signal can guarantee that itscomplete separation. To reduce this risk impact, this paper presents a novel nonnegative matricesdecomposition algorithm and the methods to discussions, and at the same time for image blindsource problems are given specific analysis and solution. The idea is to use non-negative matrixdecomposition of the sparse expression and independence data analysis, to solve the signalprocessing blind source separation problem. Through the simulation test, get a lot of real data,the analysis achieve the expected results.