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
Download full-text


Available from: Serhat Selcuk Bucak
  • Source
    • "In the same category, Yamazaki et al. [30] and Tsai et al. [46] have used an Independent Component Analysis (SL-ICA). In another way, Bucak et al. [31] [47] have proposed an Incremental Non-negative Matrix Factorization (SL-INMF) to reduce the dimension. In order to take into account the spatial information, Li et al. [32] have used an Incremental Rank-(R 1 ,R 2 ,R 3 ) Tensor (SL-IRT). "
    [Show abstract] [Hide abstract]
    ABSTRACT: Background modeling is often used to detect moving object in video acquired by a fixed camera. Recently, subspace learning methods have been used to model the background in the idea to represent online data content while reducing dimension significantly. The first method using Principal Component Analysis (PCA) was proposed by Oliver et al. [1] and a representative patent using PCA concerns the detection of cars and persons in video surveillance [2]. Numerous improvements and variants were developed over the recent years. The purpose of this paper is to provide a survey and an original classification of these improvements. Firstly, we classify the improvements of the PCA in term of strategies and the variants in term of the used subspace learning algorithms. Then, we present a comparative evaluation of the variants and evaluate them with the state-of-art algorithms (SG, MOG, and KDE) by using the Wallflower dataset.
    Full-text · Article · Nov 2009
  • Source
    • "Allili et al. (2007) [23] Yamazaki et al. (2006) [24] Bucak et al. (2007) [26] Li et al. (2008) [28] III. BACKGROUND MODELING USING TYPE-2 FGMM "
    [Show abstract] [Hide abstract]
    ABSTRACT: Gaussian Mixture Models (GMMs) are the most popular techniques in background modeling but present some limitations when some dynamic changes occur like camera jitter, illumination changes, movement in the background. Furthermore, the GMM are initialized using a training sequence which may be noisy and/or insufficient to model correctly the background. All these critical situations generate false classification in the foreground detection mask due to the related uncertainty. In this context, we propose to model the background by using a Type-2 Fuzzy Gaussian Mixture Models. The interest is to introduce descriptions of uncertain parameters in the GMM. Experimental validation of the proposed method is performed and presented on a diverse set of RGB and infrared videos. Results show the relevance of the proposed approach.
    Full-text · Article · Jan 2009
  • Source
    • "This is mainly because of the additivity property of NMF that provides bases to capture local components of the content. Our previous work on video content representation by incremental subspace learning [3] [4] have driven us to benefit from NMF in copy detection task. "
    [Show abstract] [Hide abstract]
    ABSTRACT: ITU MSPR Group participates the TREC Video Retrieval Evaluation (TRECVID) in Content Based Copy Detection (CBCD) task. The system proposed by ITU MSPR consists of two main modules: Extraction of video fingerprints and search/retrieval. We propose a feature extraction scheme based on the Nonnegative Matrix Factorization(NMF)[1], which is an efficient dimension reduction technique in video processing[2]. Video fingerprint generation module takes the factorization matrices generated by NMF as its input and converts them to binary hashes by differencial coding. Extracted hashes are indexed into a database. Searching module first applies a hash matching procedure to locate potential matching points. It is followed by temporal merging that eliminates false alarms while combining subsegments. Initial results are promising for insertion of pattern, reencoding, blurring, change of gamma and noise addition. Future work will include impoving the current results and searching for robustness to geometric transformations such as shift, crop, flip and picture-in-picture.
    Full-text · Conference Paper · Jan 2008
Show more