Conference Paper

Incremental Non-negative Matrix Factorization for Dynamic Background Modelling.

Conference: Pattern Recognition in Information Systems, Proceedings of the 7th International Workshop on Pattern Recognition in Information Systems, PRIS 2007, In conjunction with ICEIS 2007, Funchal, Madeira, Portugal, June 2007
Source: DBLP

Full-text

Available from: Serhat Selcuk Bucak, Apr 16, 2015
0 Followers
 · 
350 Views
  • [Show abstract] [Hide abstract]
    ABSTRACT: In chromatogram analysis, overlapped chromatograms are difficult to analyze if they are not resolved. The conventional multivariate resolution techniques do not give accurate results when the chromatograms are severely overlapped. In this work, ML-NMFdiv, modified non-negative matrix factorization (NMF) with divergence objective algorithm has been proposed for the separation of severely overlapped chromatograms of acetone and acrolein mixture. Before applying NMF, principal component analysis (PCA) is applied to determine number of components in the mixture taken. Most of the NMF algorithms used so far for chromatogram separation do not converge to a stable limit point and no uniqueness in the results. To get unique results, instead of random initialization, three different initialization methods namely, Robust initialization, NNDSVD (Non-Negative Double Singular Value Decomposition) based initialization and EFA (Evolving Factor Analysis) based initializations, have been used in this work and the performances are compared. The multiplicative update of already existing NMFdiv algorithm has been modified and proposed in this work as ML-NMFdiv (NMFdiv with modified multiplicative update) for overlapped chromatogram separation to improve the convergence. The proposed ML-NMFdiv algorithm is applied on the simulated and experimental chromatograms obtained for acetone and acrolein mixture. The results of proposed ML-NMFdiv are compared with existing Multivariate Curve Resolution-Alternating Least Square (MCR-ALS) method.
    Applied Mathematics and Computation 08/2014; 241:242–258. DOI:10.1016/j.amc.2014.05.022 · 1.60 Impact Factor
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Current Background Subtraction (BGS) algorithms are mostly pixel-based methods. We propose an Interest-Point(IP)-based BGS algorithm applicable in IP-based Computer Vision applications. Based on a block-wise processing strategy, the frames are divided into blocks of the same size. IPs inside each block are together Events. Throughout the frame sequence, the algorithm stores the Events in each block as well as the numbers of their occurrences (Repetition Index (RI)) in a Binary Tree. The RI is used to classify Events as either background or foreground. The background Events appear significantly more often than foreground Events. Events with an RI greater than a certain threshold are classified as background, the rest as foreground. This Event classification is used to label IPs of frames into the foreground and background IPs. Experimental results quantitatively show that the proposed algorithm delivers a good subtraction rate in comparison with other BGS approaches. Moreover, it creates a map of the background usable for further processing, it is robust to changes in illumination and can keep itself updated to changes in the background.
  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Background subtraction is one of the basic low-level operations in video analysis. The aim is to separate static information called "background" from the moving objects called "foreground". The background needs to be modeled and updated over time to allow robust foreground detection. Reconstructive subspace learning such as Principal Component Analysis (PCA) has been widely used in background modeling by significantly reducing the data's dimension. However, reconstructive representations strive to be as informative as possible in terms of well approximating the original data. On the other hand, discriminative methods such as Linear Discriminant Analysis (LDA) provides a supervised reconstruction of the data which will often give better classification results when compared to the reconstructive methods. In this paper, we offer to use and validate the combination of a reconstructive method with a discriminative one to model robustly the background. The objective is firstly to enable a robust model of the background and secondly a robust classification of pixels as background or foreground. Results on different datasets demonstrate the performance of the proposed approach.
    International Conference on Image Processing, Computer Vision, and Pattern Recognition, Las Vegas, USA,; 07/2012