Cogollos van der Linden, Jacobo Josep (2009)
Scene Classification Using a Hybrid Generative/Discriminative Approach.
Technische Universität Darmstadt
Masterarbeit, Bibliographie
Kurzbeschreibung (Abstract)
The scene classification area has been growing over the last years, becoming relevant in order to be able to work with the many digital images that are being taken every day around the world. The scope of this thesis is the scene classification task. The purpose of it is, having a group of image categories, to be able to classify new images into one of those categories. The approach done is based on the work of Bosch et al.7, using the SIFT (Scale Invariant Feature Transform) features extracted from the image applying the k-Means clustering method to have a visual vocabulary and making a quantization over the features resulting on a BoW (Bag of Words). To the BoW a hybrid approach is applied for the classification, using the pLSA(probabilistic Latent Semantic Analysis technique) and the SVM (Support Vector Machine) classification technique. The pLSA is used to decrease the dimensionality of the information that is treated and find relations between it, and the SVM for the posterior classification applying the one-vs-all rule. In this work it is shown that the concatenation of the PHOG (Pyramid of Histograms of Orientation Gradients) descriptor with the BoW leads to a better global performance. Further the inclusion of color information through histograms improved the results. The approach has been applied to different data-sets using several parameter configurations.
| Typ des Eintrags: | Masterarbeit |
|---|---|
| Erschienen: | 2009 |
| Autor(en): | Cogollos van der Linden, Jacobo Josep |
| Art des Eintrags: | Bibliographie |
| Titel: | Scene Classification Using a Hybrid Generative/Discriminative Approach |
| Sprache: | Englisch |
| Publikationsjahr: | 2009 |
| Kurzbeschreibung (Abstract): | The scene classification area has been growing over the last years, becoming relevant in order to be able to work with the many digital images that are being taken every day around the world. The scope of this thesis is the scene classification task. The purpose of it is, having a group of image categories, to be able to classify new images into one of those categories. The approach done is based on the work of Bosch et al.7, using the SIFT (Scale Invariant Feature Transform) features extracted from the image applying the k-Means clustering method to have a visual vocabulary and making a quantization over the features resulting on a BoW (Bag of Words). To the BoW a hybrid approach is applied for the classification, using the pLSA(probabilistic Latent Semantic Analysis technique) and the SVM (Support Vector Machine) classification technique. The pLSA is used to decrease the dimensionality of the information that is treated and find relations between it, and the SVM for the posterior classification applying the one-vs-all rule. In this work it is shown that the concatenation of the PHOG (Pyramid of Histograms of Orientation Gradients) descriptor with the BoW leads to a better global performance. Further the inclusion of color information through histograms improved the results. The approach has been applied to different data-sets using several parameter configurations. |
| Freie Schlagworte: | Image retrieval, Image classification, Scene classification, Bag-of-words |
| Zusätzliche Informationen: | 67 p. |
| Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Graphisch-Interaktive Systeme |
| Hinterlegungsdatum: | 12 Nov 2018 11:16 |
| Letzte Änderung: | 21 Apr 2021 15:56 |
| PPN: | |
| Export: | |
| Suche nach Titel in: | TUfind oder in Google |
 |
Frage zum Eintrag |
Optionen (nur für Redakteure)
 |
Redaktionelle Details anzeigen |
Drucken
Impressum