Reconstructive and Discriminative Sparse Representation for Visual Object Categorization

International audienceSparse representation was originally used in signal processing as apowerful tool for acquiring, representing and compressinghigh-dimensional signals. Recently, motivated by the great successes it hasachieved, it has become a hot research topic in the domainof computer vision and pattern recognition. In this paper, we propose to adapt sparse representation to the problem of Visual Object Categorization which aims at predicting whether at least one or several objects of some given categories are present in an image. Thus, we have elaborated a reconstructive and discriminative sparserepresentation of images, which integrates a discriminative term, such asFisher discriminative measure or the output of a SVM classifier, intothe standard sparse representation objective function in order tolearn a reconstructive and discriminative dictionary.Experiments carried out on the SIMPLIcity image dataset have clearlyrevealed that our reconstructive and discriminative approach has gained an obviousimprovement of the classification accuracy compared to standard SVMusing image features as input. Moreover, the results have shown that our approach is more efficient than a sparse representation being only reconstructive, which indicates that adding a discriminative term forconstructing the sparse representation is more suitable for thecategorization purpose