Automatic separation of skin lesions from healthy human skin

Geertsema, S.L. and Feringa, S. (2012) Automatic separation of skin lesions from healthy human skin. Bachelor's Thesis, Computing Science.

Preview	Text INF-BA-2012-S.L.Geertsema-S.Feringa.pdf - Published Version Download (16MB) | Preview
	Text GeertsemaAkkoord.pdf - Other Restricted to Registered users only Download (13kB)

Abstract

This bachelor thesis presents an implementation of a skin lesion segmentation system in digital images using Learning Vector Quantization (LVQ) that potentially could be used by physicians for skin lesion classification systems. A feature vector is extracted with different feature types including mean colour value and its standard deviation value, histograms and edge detection using two different colour models. For all images in the training dataset for use by LVQ, two subimages of an image with a skin lesion are taken, one subimage shows sick skin, one shows healthy skin. New images are divided up into clusters and each cluster is being compared in the classification phase of LVQ. This comparison is done between each clusters' feature vector and the feature vectors of the prototypes in the training set that have been made with the prototype generation phase of LVQ. A neighbourhood filter is also tested as a post-processing tool. F-scores are used as a result measurement and the maximum mean F-score produced by this software is for a dataset of 40 Melanoma images - with the neighbourhood filter not applied - a score of 0,79 for K = 5

Item Type:	Thesis (Bachelor's Thesis)
Degree programme:	Computing Science
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	15 Feb 2018 07:50
Last Modified:	15 Feb 2018 07:50
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/10380

Actions (login required)

View Item