On an optimization model for Approximate Nonnegative Matrix Factorization



Within image and video signal processing several tasks must be performed in order to obtain an efficient and low-dimensional encoding for transmission. Studies have shown that Nonnegative Factorization of the data Matrix (NFM) is a suitable tool for achiving ranking reduction and sparse coding. Thus, developing efficient optimization algorithmic methods for video/image encoding presents a key technical challenge in the effort to minimize distortion and obtain good quality reduced encodings. In this paper, we propose a discrete optimization framework where NMF is formulated as a nonlinear programming optimization problem, and apply an Spectral Projected Gradient (SPG) algorithm to this problem. The empirical experience indicates that the proposed approach can in fact produce good quality local minima.


Nonnegative Matrix Factorization; Constrained Large Scale Optimization; Image Signal Processing;


Signal Processing; Feature Extraction

Related Project


Book Chapter

Computational Intelligence and Decision Making: Trends and Applications, Eds: A.Madureira, C.Reis, and V. Marques, 23, pp. 249-257, Springer - Intern Series on Intelligent Systems Control. And Automation Science And Engineering, January 2013

Cited by

Year 2013 : 1 citations

 Tomé, A.M. and Schachtner, R. and Vigneron, V. and Puntonet, C.G. and Lang, E.W., "A logistic non-negative matrix factorization approach to binary data sets", Multidimensional Systems and Signal Processing, 1-19, 2013, Springer US.
DOI: 10.1007/s11045-013-0240-9