Analysis of Convolutional Neural Networks for Detection and Classification of Damages in Fingerprint Images

The aim of this Master's thesis is to analyze detection and classification approaches using convolutional neural networks on the problem of fingerprint damage. The first part of the thesis deals with the study of literature related to biometrics and fingerprint processing with emphasis on possible diseases that may affect the fingertip area. Subsequently, the thesis focuses on neural network-based recognition. The thesis describes the architectures of convolutional neural networks and object detection approaches up to the latest research. Several detection methods for detection and classification of skin diseases affecting fingertip are proposed using modern architectures, different types of backbone networks and detection methods. Eight models based on four different detection and classification approaches are chosen for the experiments. Subsequently, each model is trained several times using configuration parameter adjustments. The models are assessed on the basis of various metrics and compared in terms of the use of the backbone network and the chosen method for detection. The best result of 76.875 % was achieved in the test of correctly detected and classified area on real fingerprint images. The most problematic disease for detection and classification was atopic eczema, whose symptoms can manifest in many ways