Superpixel-based online wagging one-class ensemble for feature selection in foreground/background separation
In the last decades, researchers in the field of Background Subtraction (BS) have developed methods to handle the different type of challenges. However, at the present time, no traditional algorithm seem to be able to simultaneously address all the key BS challenges. This can mainly be attributed to the lack of systematic investigation concerning the role and the importance of features within background modeling and foreground detection. In this work, we present a novel online one-class ensemble based on wagging to select suitable features to each region of a certain scene to distinguish the foreground objects from the background. In addition, we propose a mechanism to update the importance of each feature discarding insignificantly features over time. The experimental results on three challenging datasets (i.e MSVS, RGB-D object detection, CD.net 2014) show the pertinence of the proposed approach.
HIGHLIGHTS
* A novel methodology to select the best features based on wagging.
* A superpixel segmentation strategy to improve the segmentation performance, increasing the computational efficiency of our ensemble.
* A mechanism called Adaptive Importance Computation and Ensemble Pruning (AIC-EP) to suitably update the importance of each feature discarding insignificantly features over time.
* A superpixel segmentation strategy to improve the segmentation performance, increasing the computational efficiency of our ensemble.
* A mechanism called Adaptive Importance Computation and Ensemble Pruning (AIC-EP) to suitably update the importance of each feature discarding insignificantly features over time.
BRIEF OVERVIEW OF THE PROPOSED FRAMEWORK
Fig.1. Brief overview of the proposed framework. A set of features are extracted from the training image sequence. Next, our wagging version creates different pools of IWOC-SVM classifiers from a certain feature. A heuristic approach called Small Votes Instance Selection (SVIS) is used in the IWOC-SVM model updating step. Finally, we use a mechanism called Adaptive Importance and Ensemble Pruning (AIC-EP) to update the importance of the classifiers discarding insignificantly classifiers over time. Only the classifiers with high importance are selected and combined to form a strong classifier. The whole framework described here works as incremental manner.
ALGORITHM: THE WAGGING FOR FEATURE SELECTION
EXPERIMENTAL RESULTS
Background subtraction results on MSVS dataset
Fig.2. Background subtraction results using the MSVS (top row), RGB-D (middle row) and {CD}net 2014 (bottom row) datasets -- (a) original frame, (b) ground truth and (c) proposed method. The true positives (TP) regions are in white, true negatives (TN) regions in black, false positives (FP) regions in red and false negatives (FN) regions in green
Background subtraction results on RGB-D dataset
Fig.3. Background subtraction results on RGB-D dataset – (a) original frame, (b) features map showing most important feature for each region and (c) its respective histogram of features importance.
PERFORMANCE OF THE DIFFERENT METHODS USING THE RGB-D DATASET
PUBLICATION AND SOURCE CODE
2017 - Silva, C. and Bouwmans, T. and Frélicot, C. “Superpixel-based incremental wagging one-class ensemble for feature selection in foreground/background separation”. Pattern Recognition Letters (PRL). [PDF] [CODE]
2016 - Bouwmans, T. and Silva, C. and Marghes, C. and Zitouni, S. and Bhaskar, H. and Frélicot, C. “On the Role and the Importance of Features for Background Modeling and Foreground Detection”. Computer Science Review, 2016. [PDF]
