Marco Ross: Computer aided analysis of ancient brick walls

This report is about a method to automatically segment colour images of brick walls, in order to classify regions as brick or mortar material. Furthermore geometrical features of bricks shall be extracted. A support vector machine is used to perform a pixel-based classification of colour images and standard statistical methods are applied to extract the requested features from the segmented regions. During the experimental phase different approaches (e.g. Support Vector Machines, watershed segmentation and morphological operations) were followed for the diffcult task of segmentation, since the quality of the segmentation is critical to the whole process. In general the pixel-based SVM classifier performed out best. As a bonus the SVM also was the computationally least expensive variant that was tested. A linear kernel function is used, so the classification can be done by a simple perceptron evaluation. For future applications it is proposed to exhaust the ease of computation by applying an online perceptron training during manual post-processing.

What	Presentation
When	Nov 04, 2008 from 04:00 pm to 05:00 pm
Where	SEM 183-2
Add event to calendar	vCal iCal