@MASTERSTHESIS\{IMM2013-06684,
    author       = "I. Danov",
    title        = "Block-wise Finger Image Quality Assessment Based on Machine Learning",
    year         = "2013",
    school       = "Technical University of Denmark, Department of Applied Mathematics and Computer Science",
    address      = "Matematiktorvet, Building 303B, {DK-}2800 Kgs. Lyngby, Denmark, compute@compute.dtu.dk",
    type         = "",
    note         = "{DTU} supervisor: Rasmus Larsen, rlar@dtu.dk, {DTU} Compute",
    url          = "http://www.compute.dtu.dk/English.aspx",
    abstract     = "The development of fingerprint quality assessment algorithms has seen increasing popularity throughout the recent years, since the accuracy of the fingerprint biometric systems is heavily dependent on the quality of the acquired samples. There exist a variety of image analysis methods for quality estimation, however they are usually characterized with high computational complexity. Block-wise finger image quality assessment based on Self-Organizing Maps (SOM) is a novel approach that has shown promising results in terms of speed and performance.
The goal of the thesis is to conduct experiments by training {SOM} networks with a large dataset of raw fingerprint image blocks in order to extract quality features. These features are to be interpreted by another machine learning model trained to predict the quality score of the fingerprint image. In this thesis two datasets with two block sizes each are used for training linearly initialized {SOM} networks.
The results from the trainings are analyzed and the extracted quality features are used as an input for the training of four different machine learning models. These four machine learning models are used to predict the quality scores of the fingerprint images and are based on three machine learning techniques: Self-Organizing Maps, Generative Topographic Mapping and Random Forests. The performance of the models is comparatively evaluated with two state of the art approaches using {ERC} curves and Spearman’s Correlation matrices.
Some of the proposed methods show improvement of the results obtained by previous work in the field and can possibly take part of the {ISO}/{IEC} or {NIST} Finger Image Quality 2.0 standards."
}