Home Publications
 
Home Publications
Background modeling through dictionary learning
Year: 2013 Keywords: background modeling, dictionary learning, sparse coding
Authors: Alessandra Staglian√≤, Nicoletta Noceti, Alessandro Verri, Francesca Odone  
Book title: IEEE International Conference on Image Processing
   
Abstract:
n this work we build a model of the background based on dictionary learning. The image is divided into patches of equal size and a background model is obtained as a sparse linear combination of patch prototypes learnt from the im- age stream. In the bootstrap phase, the dictionary consists of the first few frames. The fact that many very similar image patches are fed as examples into the dictionary learning mod- ule ensures that, for each patch, the dictionary grows adap- tively in size only when the reconstruction accuracy is not sufficient. By enforcing sparsity, the obtained reconstruc- tion can also be computed and maintained effectively. Im- age variations not leading to stable changes do not trigger an update of the current dictionary and are discarded. As a result the proposed method is stable with respect to illu- mination changes, correctly incorporates stable background changes in the model, and cancels out moving objects. Exper- iments on benchmark data indicate that the proposed method reaches very good pixel-wise performances even if relatively large patches are used. The devised model, fully data driven, appears to be a suitable as a main component of a change detection system.