VSE knjižnice (vzajemna bibliografsko-kataložna baza podatkov COBIB.SI)
  • Local probabilistic descriptors for image categorisation
    Mele, Katarina ; Šuc, Dorian ; Maver, Jasna
    Image categorisation involves the well known difficulties with different visual appearances of a single object, but also introduces the problem of within-category variation. This within-category ... variation makes highly distinctive local descriptors less appropriate for categorisation. Difficulties because of the within category variation and clutter are tackled by modelling image fragments in a new manner. The authors propose a family of local image descriptors, called probabilistic patch descriptors (PPDs). PPDs encode the appearance of image fragments as well as their variability within a category. PPDs extend the usual local descriptors by also modelling the variance of the descriptors' elements, for example pixels or bins in a histogram. To compare two PPDs, and a PPD with an image, a new similarity measure called PPD matching score is introduced. For each object category, a set of representative PPDs is learnt. Images are represented as feature vectors of the best matching scores obtained for representative PPDs in images. Support vector machine classifiers are then trained on the feature vectors. PPDs are applied to image categorisation using machine learning where the features are the matching scores between images and PPDs. The authors experiment with two variants of PPDs that are based on complementary local descriptors. An interesting observation is that combining the two PPD variants improves the accuracy of categorisation. Experiments indicate that the benefits of modelling the within-category variation give results that are comparable with the state-of-the-art categorisation methods, and show good robustness with respect to noise and occlusions.
    Vir: IET computer vision. - ISSN 1751-9632 (Vol. 3, 1, 2009, str. 8-23)
    Vrsta gradiva - članek, sestavni del ; neleposlovje za odrasle
    Leto - 2009
    Jezik - angleški
    COBISS.SI-ID - 38466658

Vnos na polico