Téma disertační práce

Školitel:Zbořil František V., doc. Ing., CSc.
Téma:Rozpoznávání obrazů
Zahájení v ak.r.:1996/1997
Obhajoba disertace:2000-12-11
Název disertace:Rozpoznávání obrazu se zaměřením na identifikaci osob dle otisku prstu
Charakteristika řešeného problému:

Fingerprint images are often used for the identification of individuals. The aim of this work is to find an appropriate method for the fingerprint image recognition. The method should be able to verify whether two fingerprint images belong to the same finger. The fingerprint image recognition is often based on the extraction of characteristic features (usually minutiae) from pre-processed images. Extracted characteristic features are subsequently compared. Pre-processing of the fingerprint images usually includes filtration, segmentation, smoothing, thinning and additional modifications. Unfortunately pre-processing introduces errors - false minutiae can appear, the minutiae are shifted off the expected position, etc. The elasticity of the skin causes deformations in fingerprint images. These errors influence the efficiency of the subsequent recognition. The authors of different methods try to compensate the errors caused by pre-processing in different ways. Here, a method capable to align and recognize two fingerprint images directly from the grey-scale images is designed. Fingerprint images are aligned by means of mutual information. No pre-processing is needed. Images are divided into parts (triangles) that are gradually aligned. The image inside each triangle is adapted so that the mutual information between the two corresponding triangles is maximized. The method is able to compensate the deformations caused by the elasticity of the skin with a sub-pixel precision. Affine transformations are used to achieve the best geometrical alignment of the images. The applied method gives promising results. The advantage of this method is that no pre-processing is needed. The method is able not only to decide whether two images belong to the same finger, but also to return adapted and aligned images. Another advantage is that the method is general and it can be applied to other types of images than fingerprints.

Publikace související s vypsaným tématem:
2000ZBOŘIL František a ZBOŘIL František ml. The use of the RCE Neural Network in a Pattern Recognition. In: 34th Spring International Conference Modelling and Simulation of Systems MOSIS 2000. Ostrava: MARQ, 2000, s. 65-70. ISBN 80-85988-44-5.

Vaše IPv4 adresa: 35.175.191.168
Přepnout na https