|
Abstract:
|
In this thesis, the performances of a content-based indexing and retrieval framework are analyzed, when various levels of Gaussian blur distortion are introduced. For the experiments the MUVIS, Multimedia Video Indexing and Retrieval System framework, developed at Tampere University of Technology, Institute of Signal Processing, is used. This framework allows algorithm implementation, testing, configuration, and comparing for capturing, recording, indexing and retrieval purposes, combined with browsing and various other visual and semantic capabilities. The analysis of different features extraction modules and different image databases is thus possible. The aim of this work is to analyze the effects of blur introduction in large and medium image databases, with the purpose to understand how it is possible to maintain good retrieval performance although working with blurred images. The analysis of two medium resolution image databases is performed. The first database contains 10.000 natural images from diverse contents such as wild life, city, buses, horses, mountains, beach, food, African natives, etc., divided in 100 classes depending on the content. The second database contains 300 images, original and degraded, extracted from the previous one and divided in 6 classes depending on the content. In each class there are original images and different degraded versions of them, according to the amount of the introduced Gaussian blur distortion. Considering several feature extraction modules, the capacity of the system to retrieve relevant images from the collection is analyzed, both in the cases of degraded images collection and degraded query images. |