PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Application of Self-Organizing Maps and automatic image segmentation to 101 object categories database
Jorma Laaksonen, Ville Viitaniemi and Markus Koskela
In: Fourth International Workshop on Content-Based Multimedia Indexing, 21-23 Jun 2005, Riga, Latvia.


In this paper, we study how well our PicSOM CBIR system is able to find prototypical image segments based on image-level keywords and automatic image segmentation. We also study different methods for focusing a given keyword on a particular image segment. Both these processes are based on the Self-Organizing Map's ability to map image segments which are mutually similar according to a specific image feature in nearby map units. In addition, the PicSOM system can automatically use and weight multiple different features in parallel. In the automatic image segmentation applied to the images of the 101 Object Categories database, a fixed number of segments have been extracted from each image. This leads in many cases to oversegmentation, but our experiments show that the system's ability to find prototypical segments is not severely impaired. On the other hand, it is clear that the process of keyword focusing would benefit from more precise segmentations.

PDF - Requires Adobe Acrobat Reader or other PDF viewer.
EPrint Type:Conference or Workshop Item (Paper)
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
Machine Vision
Theory & Algorithms
Information Retrieval & Textual Information Access
ID Code:1727
Deposited By:Jorma Laaksonen
Deposited On:28 November 2005