Background Subtraction Research

My research in background subtraction concerns  full surveys in the field for novice and experts, investigation of mathematical concepts, machine learning concepts and signal processing concepts to model the background, and classify pixels as background or foreground, and robust features and feature selection for background/foreground separation. Thus, my main research can be summarized in the following points.

1. Surveys

• • A Survey Approach: The surveys concern sub-categories of models (Mixture of Gaussians, Subspace Learning, RPCA), categories of models (Fuzzy, Decomposition in Low-rank+ Additive Matrices, Statistical), and all the categories of models. (6 Chapters, 6 Journals)

2. Mathematical Concepts

• • A Statistical Approach:  This research concerns the improvement of the most investigated method in statistical background modeling, that is the Mixture of Gaussians (MOG). (1 Conference)

  • A Fuzzy Approach: This research introduces fuzzy concepts in the different steps of background subtraction. Type-2 Fuzzy Gaussian is used for background modeling, Choquet Integral is used in foreground detection, An adaptive fuzzy scheme is used for background maintenance via a fuzzy  adaptive learning rate. (1 Chapter, 1 Journal, 6 Conferences)

3. Machine Learning Concepts

3.1 Regular Subspace Learning

• • A Subspace Learning Approach: This research concerns the use of discriminative and mixed subspace learning in background modeling and foreground detection. Thus, IMMC is used as a discriminative approach and PCA-LDA is used as a mixed approach. (2 Journals, 3 Conferences)

3.2 Robust Subspace Learning/Dynamic Subspace Learning

• • A Decomposition into Low rank plus Additive Matrices Approach: This research firstly concerns the evaluation of the RPCA matrix model for background/foreground separation and background initialization. Secondly, it investigates the application of similar problem formulations (RPCA, RNMF, RMC, RST, RLRM) based on decomposition into low rank plus additive matrices. (3 Chapters, 6 Journals, 17 Conferences)

  • A Decomposition into Low rank plus Additive Tensors Approach: This research concerns the evaluation of the RPCA tensor model for background/foreground separation and background initialization. Then, we introduce online decomposition into low rank plus additive tensors for background/foreground separation and background initialization. (4 Conferences)

3.3. Deep Learning

• • A Deep Learning Approach : This research concerns the evaluation of deep learning models for background/foreground separation, the application of the supervised DeepSphere model, and the unsupervised GAN model. (2 Journals, 6 Conferences)

  • A Graph Neural Networks : This research concerns the evaluation of transductive and inductive GNNs for background/foreground separation. (2 Chapters, 1 Journal, 2 Conferences)

4. Signal Processing Concepts

• • A Graph Signal Processing Approach: This research proposes to employ Graphs Signal Processing concepts for background/foreground separation. (1 Chapter, 1 Journal, 3 Conferences)

5. Features

• • Robust Features and Feature Selection: This research concern the development of robust texture features and feature selection for robust background/foreground separation. (2 Journals, 3 Conferences)

Publications (76) : 10 (13-3) chapters, 21 Journals, 45 conferences.

Note: My publications are available on Academia, Research Gate, Researchr, ORCID and Publication List.