Adaptive segmentaion and similarity measures in content-based image retrieval

• Main Author: Ricky Purnomo.
• Other Authors: Chan, Kap Luk
• Format: Theses and Dissertations
• Language: English
• Published: 2008
• Subjects: DRNTU::Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision; DRNTU::Engineering::Computer science and engineering::Information systems::Information storage and retrieval; DRNTU::Engineering::Computer science and engineering::Information systems::Database management
• Online Access: http://hdl.handle.net/10356/2433
• Institution: Nanyang Technological University

With the advances in digital imaging, the accompanying increase in the number of digital images, and subsequent creation of image databases in digital libraries, image retrieval has emerged as a problem which merits some attention. This thesis aims to investigate particularly natural (general) image retrieval. A short review of works on image retrieval is presented and followed by description of the colour and texture features used in this project, the L*a*b* colour space and the construction of Gabor filter bank. A new adaptive image segmentation method based on Fuzzy C-Means is developed for region extraction from an image. This algorithm is developed to support region based image retrieval. Several methods used in the segmentation algorithm for finding the most accurate number of regions are evaluated, with the one using fast merging Davies Bouldin index giving a reasonably good result, trading off accuracy for speed. Global and region based image retrieval methods using Nearest Neighbour distance, modified Nearest Feature Line distance, and Region Match Distance are investigated. While all of them yields relatively good results, some performance characteristics unique to certain combination of image types/classes and methods are analyzed.