Javascript must be enabled for the correct page display

Learning scale invariant filters

Boersma, N. J. (2004) Learning scale invariant filters. Master's Thesis / Essay, Computing Science.

[img]
Preview
Text
Infor_Ma_2004_NJBoersma.CV.pdf - Published Version

Download (1MB) | Preview

Abstract

Connected filters are image operators that work on the connected components (in the binary case), or flat zones (in the grey-scale or color case) of an image. We have developed a connected filter which can be trained to respond to certain types of shapes. Such a filter can for example be used for image recognition purposes. A mathematical representation of the shape of an object is used in the form of moments. Since the shape of an object does not change under translation, rotation and scaling, the moments have to be invariant against these transformations. A Max-Tree algorithm is used to extract the connected components from an input image such that the shape ifiter can be applied to them individually. A standard Max-Tree algorithm is adapted, such that it implements a nearest neighbor classifier which can handle vector attributes like moments. In this report, we will present implementation details and a comparison between two different shape descriptors: Hu moment invariants (1962) and Krawtchouk moment invariants (2003). All insights that were gathered during the project are presented as well.

Item Type: Thesis (Master's Thesis / Essay)
Degree programme: Computing Science
Thesis type: Master's Thesis / Essay
Language: English
Date Deposited: 15 Feb 2018 07:30
Last Modified: 15 Feb 2018 07:30
URI: http://fse.studenttheses.ub.rug.nl/id/eprint/8890

Actions (login required)

View Item View Item