Beta 1


Title Multispectral Textile Analysis
Author Petersen, Anders
Supervisor Carstensen, Jens Michael (Image Analysis and Computer Graphics, Department of Informatics and Mathematical Modeling, Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark)
Institution Technical University of Denmark, DTU, DK-2800 Kgs. Lyngby, Denmark
Thesis level Master's thesis
Year 2008
Abstract This thesis describes how the use of a patented camera from Videometer A/S, can be the foundation for a range of quality control measures on textiles. An algorithm for segmenting the threads from denim will be tested on different fabrics, where abrasion and backstaining will be measured and compared to proven measuring methods done with a chromameter. New features for denim will also be presented such as speckle and graininess, which will give an objective measure of the global contrast in the fabric. Two methods for estimation of fuzz & pills on knitted fabric will be evaluated. One uses fourier analysis to remove the repeated knitted pattern by filtering in the fourier domain with a automated generated filtering mask, and thereby makes a fuzz & pills estimate from the fourier domain. The other method uses gaussian pyramids to limit the repeated knitted pattern, where after a ratio of the difference in the variation will provide information of the amount of fuzz & pills. A third approach will not directly estimate the amount of fuzz & pills, but instead use a linear regressionmodel, which will predict the amount of fuzz & pills from a set of features. Different features will be evaluated, and combined with a step-wise forward selection scheme, and be compared with principal component regression and ’best subset’ selection. Over 100 GB of images have been collected to test these different algorithms strength and weaknesses for daily situations for an enzyme company Genencor.
Series IMM-M.Sc.-2008-04
Fulltext
Original PDF ep08_04_new_pdf.pdf (9.80 MB)
Admin Creation date: 2008-03-12    Update date: 2008-07-15    Source: dtu    ID: 211665    Original MXD