@MASTERSTHESIS\{IMM2010-06000,
    author       = "J. K. Janot",
    title        = "Modellering af fugtigheden i br{\o}d p{\aa} baggrund af multispektrale billeder",
    year         = "2010",
    month        = "dec",
    keywords     = "Fugtighed i b{\o}rd, multispektrale billeder",
    school       = "Informatics and Mathematical Modelling, Technical University of Denmark",
    address      = "",
    type         = "",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/6000-full.html",
    abstract     = "This thesis is motivated by a desire to analyse changes in the spatial distribut
ion of moisture in bread, since moist redistribution is related to the loss
 of freshness, i.e. the quality of the bread.

Thus the thesis presents a method for estimating the distribution of moist
 in bread on the basis of multispectral images.
 The multispectral images consist of 18 spectral bands within the range
 of ultrablue (450 nm) and nearinfrared (970 nm).
 For the application in question, it is desirable to be able to determine
 the moist distribution in the bread in a reliable, inexpensive and effective
 manner.

When enzymes and emulsifiers are added to the bread-mix, loss of freshness
 can be slowed down, thus the method presented here, may be useful as a
 quality inspection tool, when adjusting the concentrations of the additives.

The thesis' main focus is feature extraction from textures and multivariate
 regression with variable selection on high dimensional datasets.

The thesis explains how the images can be divided into small regions, with
 the purpose of obtaining more observations and how we extract features
 from the marginal distribution and the co-occurrence matrix of the texture
 in each of the regions.
 The areas in the image where within the regions are placed, corresponds
 to the test pieces, on which the moisture content has been measured.
 The reference method is; desiccation ofthe water in an oven.
 Thus the model is based on features extracted from regions of multispectral
 image and the corresponding measurements of moist.

The thesis examines the usability of different statistical methods such
 as linear regression, regression trees with a splines function basis, and
 ensembles of regression trees with a constant function basis.
 During this model selection part of the thesis, we will also consider different
 region sizes and various feature types.

In the final part of the report, we present a novel method to obtain high
 resolution images of the distribution of moist in the images.
 And we show how this can be used to study the process of moist redistribution
 in breads with different additives, which have been stored for 15 days.",
    isbn_issn    = "IMM-Master-2010 Thesis-95"
}