On Application Of NMF In The Recognition Of The Mobile Phone Screen

The mobile phone screen recognition, as the performance of automatic monitoringsoftware system development project, has the very high value of research and application.The actual project has very high demand on the recognition speed (single picturerecognition time is controlled in the range of10ms), the recognition speed of the traditionalalgorithm is to be further improved. Non-negative matrix factorization is proposed byD.D.Lee and H.S.Seung in1999"Nature" on all the elements in the matrix are thedecomposition method of matrix under nonnegative constraint, the method is simple andfast. The decomposition results can be explained and the storage space is small. It hasmany other advantages as well.In this paper, using non-negative matrix factorizationalgorithm for mobile phone screen image recognition. It improves the recognition speedeffectively and it also guarantees the recognition rate.This paper first introduces the basic theory of image matching, based on the commongray and feature-based image matching algorithm; then it introduces the basic principleand algorithm of non-negative matrix factorization algorithm; emphasis is put on themobile phone screen images on the MATLAB2010a platform matching simulation test:first on the training samples are non-negative matrix factorization and feature extraction,feature based training samples; then the mobile phone screen training samples and testingsamples were to feature projection, do SVM training on the projection coefficientcomputation, the recognition rate and recognition speed; and the results are compared withthe normalized cross correlation algorithm.From the experimental results, a single mobile phone picture recognition time iscontrolled in the range of10ms, and the recognition rate is close to100%; non-negativematrix factorization algorithm is better than the effect of the normalized cross-correlation.