@MASTERSTHESIS\{IMM2014-06809,
    author       = "M. J. Emerson",
    title        = "Novelty Detection of Foreign Objects Using Grating-Based Interferometry",
    year         = "2014",
    school       = "Technical University of Denmark, Department of Applied Mathematics and Computer Science",
    address      = "Richard Petersens Plads, Building 324, {DK-}2800 Kgs. Lyngby, Denmark, compute@compute.dtu.dk",
    type         = "",
    note         = "{DTU} supervisor: Line Katrine Harder Clemmensen, lkhc@dtu.dk, {DTU} Compute",
    url          = "http://www.compute.dtu.dk/English.aspx",
    abstract     = "Quality assurance in food industries is essential, both in regards to consumer satisfaction and also food safety. During food processing, unwanted foreign objects can be introduced to food products, which can both be unappetizing and hazardous for the consumer. Nowadays, {X-}ray conveyor belt solutions can detect non-organic materials, but finding organic foreign objects in food with typical {X-}ray systems is not a simple task.
The goal of the thesis is to demonstrate the improvement introduced in foreign body detection by a new {X-}ray imaging technique when organic materials are potential foreign bodies. This novel {X-}ray technique is based on interferometry, created by adding gratings to a conventional {X-}ray source. This technique provides information about a sample's absorption, refraction and scattering properties; whereas conventional {X-}rays just grant the absorption prole of a sample.
A grating-based interferometer set-up is available at Technische Universit{\"{a}}t München, where data was acquired personally. Each image, consisting of three imaging modalities (absorption, phase contrast and dark-field), contains a food sample contaminated by different sized foreign bodies (organic and non organic). Several food products were imaged. These products have dierent properties and are of importance to the {NEXIM} project (New {X-}ray Imaging Modalities for safe and high quality food) collaborators.
In this thesis, the performance of two classification algorithms is compared, one is supervised and the other is unsupervised. Focusing on the unsupervised technique, food models with varying number of features are compared and detection results are contrasted with those obtained when using only the absorption."
}