An automatic fingerprint classification technique based on singular points and structure shape of orientation fields

Abstract

Generally, an automatic fingerprint classification system aims to classify the fingerprints into several categories based on global features such as ridge structure and singular points. Its process basically covers: segmentation, enhancement, orientation field estimation, singular point detection, and classification. However, its performance is heavily relied on image quality that comes in various forms such as low contrast, wet, dry, bruise, cuts, stains, etc. Although a great effort has been made by previous studies to come out with various methods, their performances especially in terms of accuracy are fallen short, and room for improvements is still wide open. Thus, this thesis proposes an automatic fingerprint classification scheme based on singular points and structural shape of orientation fields. This method begins with foreground extractions using a composite method which combines local mean values of the grey-levels with local variances of the gradient magnitudes. Then, noise patches in the foreground are detected using coherence, and are enhanced using minimum variance of gradient magnitude. Next, Least Mean Square algorithm is applied to estimate the orientation fields, and a corrective procedure is performed on the false ones using minimum variance of the orientation fields. Later, an orientation image is created, and then partitioned into several distinct regions of homogenous orientation fields. The convergence point of these regions implicitly reveals an area that most likely contains a singular point. Subsequently, core and delta in this localized area are then detected using the Poincaré index. Finally, based on the number of core and delta and their locations, an exclusive membership of the fingerprint can be ascertained. Should it fail, the structure shape of the orientation fields neighbouring the core or delta is analysed. The performance of the proposed method is evaluated and tested using 27,000 fingerprints of NIST Special Database 14, which is considered de facto standard dataset for development and testing of fingerprint classification systems. The results obtained are very encouraging with accuracy rate of 89.31% that markedly outperformed the latest work of the renowned researchers.