Bag-of-word based brand recognition using Markov clustering algorithm for codebook generation

In order to address the issue of counterfeiting online, it is necessary to use automatic tools that analyze the large amount of information available over the Internet. Analysis methods that extract information about the content of the images are very promising for this purpose. In this paper, a met...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:2015 IEEE International Conference on Image Processing (ICIP) S. 3315 - 3318
Hauptverfasser: Benezeth, Yannick, Bertaux, Aurelie, Manceau, Aldric
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IEEE 01.09.2015
Schlagworte:
Clustering algorithms
Feature extraction
Image recognition
Markov processes
Object recognition
Symmetric matrices
Visualization
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In order to address the issue of counterfeiting online, it is necessary to use automatic tools that analyze the large amount of information available over the Internet. Analysis methods that extract information about the content of the images are very promising for this purpose. In this paper, a method that automatically extract the brand of objects in images is proposed. The method does not explicitly search for text or logos. This information is implicitly included in the Bag-of-Words representation. In the Bag-of-Words paradigm, visual features are clustered to create the visual words. Despite its shortcomings, k-means is the most widely used algorithm. With k-means, the selection of the number of visual words is critical. In this paper, another clustering algorithm is proposed. Markov Cluster Algorithm (MCL) is very fast, does not require an arbitrary selection of the number of classes and does not rely on random initialization. First, we demonstrate in this paper that MCL is competitive to k-means with a number of cluster experimentally selected. Second, we show that it is possible to identify brand from objects in images without previous knowledge about visual identity of these brands.
DOI:10.1109/ICIP.2015.7351417