Research On Fingerprint Image Orientation Field Extraction And Enhancement

The technology of biometric identification has become a hot topic in the field of information science and technology.It is a general framework for identifying biometrics-based identification system by using information processing technology.It is the organic combination of biotechnology and information technology.Biometric identification technology overcomes the shortcomings of many personal identification methods,and has high recognition accuracy and convenient use.Among all the biometrics,the fingerprint recognition technology is one of the most popular technologies for personal identification because of its convenience,uniqueness and invariance.Although the automatic fingerprint identification system(AFIS)has been extensively studied and has performed well,it has long been a challenging problem for recognizing low quality fingerprints,which are degraded by various factors such as dirt,scar,greasing and moisture on the surface of fingertips.In this paper,we focus our research on the low quality fingerprint image orientation field extraction and enhancemen.The specific research contents are as follows:1.Study the orientation field extracting(FOF)for low quality fingerprint images.In practice,the fingerprint ridge has an obvious orientation pattern;the estimation of the FOF is a very important step in the automatic fingerprint identification system(AFIS).Conventional gradient based methods are popular but very sensitive to noise.This paper presents an improved FOF extraction method.The basic idea of the algorithm is to reconstruct the FOF by combining weighted linear project analysis with orientation diffusion.First,in order to effectively remove the noise,the point orientations are fitted by using 2D discrete orthogonal polynomial.In the second procedure,the role of the gradient modulus is taken into full account,and the weights of the point orientations are obtained by computing the similarity of the fitted point orientations,and then the block orientation are estimated by the weighted linear projection analysis based on the vector set of point gradients.In the end,the FOF is reconstructed based on quality grading scheme and composite window strategy.The proposed method does not need any prior knowledge of singular points.The proposed method can ensure that the cluster of blocks with the higher quality will be estimated earlier,and the accuracy and reliability of the non-estimated block orientations with lower quality can be improved by the feedback of the estimated block orientations with higher quality.2.Study the fingerprint image enhancing based on band-pass filtering.How to improve the performance of band-pass filter is very important to fingerprint image enhancing.In order to improve the performance of band-pass filter,2D adaptive Chebyshev band-pass filter(ACBF)with orientation-selective,a novel fingerprint enhancement filter,is designed in this paper.The fingerprint enhancement is deeply rooted in the spectra diffusion by performing the 2D ACBF with orientation-selective in the frequency domain.The process of the enhancement is to have two phases: fingerprint is first enhanced by using Gabor filter and histogram equalization,and then the pre-enhanced fingerprint is enhanced based on spectra diffusion by using the 2D ACBF with orientation-selective in frequency domain.In the first stage,the fingerprint quality can be improved in some extent.In the second stage,first,the qualities of patches are evaluated by the coherence of point orientations.Second,the adaptive parameters of the 2D ACBF are estimated,further,the 2D ACBF with orientation-selective is designed.Finally,the fingerprint image is enhanced based on spectra diffusion by using the 2D ACBF with orientation-selective.3.Study the fingerprint image enhancing based on deep Boltzmann machine(DBM).In order to make up for the shortcomings of the traditional fingerprint enhancement,this paper proposes a novel algorithm by using orientation Gaussian band-pass filter(OGBPF)to enhancing the fingerprint in first,and then the DBM with orientation selection is employed to reconstruct these regions that are enhanced incorrectly in the first phase.The fingerprint is enhanced based on the quality grading scheme and the composite window strategy.In proposed method,the traditional enhancement method and deep learning method complement one another perfectly.Due to the ridge prior information is taken into full account,a unique DBM model is trained for fingerprint blocks with different orientation.The corresponding DBM model is selected for error enhancement block reconstruction,which makes the reconstruction result more stable and reliable.4.Study the fingerprint image enhancing based on dictionary learning via sparse representation.In order to overcome the shortcomings of the Gabor filter,motivated by the recent success of sparse representation in image processing,we propose a novel fingerprint enhancement method that combines Gabor filter and classification dictionaries learning.The process of the enhancement is divided into two phases: in the first stage,the training patches are classified into eight groups based on their own orientations,and the corresponding classification dictionaries are learned in frequency domain.In the second stage,fingerprint is first enhanced using Gabor filtering.The pre-enhanced fingerprint is further enhanced based on quality grading scheme and composite window strategy by using sparse representation with the priori information of ridge pattern based on classification dictionaries learning.The combination of quality grading scheme and composite window strategy promises to ensure that spectra diffusion is successfully applied.This further improves the quality of enhanced fingerprint.5.Study the fingerprint image super resolution using sparse representation by classification coupled dictionaries.This paper presents a new algorithm for reconstructing the fingerprint super-resolution image.The basic idea of the algorithm is to reconstruct the SR image by using sparse representation with ridge pattern prior based on classification coupled dictionaries.First,the orientations of training patches are estimated by the weighted linear projection analysis.In the second procedure,the qualities of patches are assessed by the coherence of point orientations,the training patches are subsequently classified into eight groups based on their own orientations and qualities,and then the training patches of each class are selected from candidate patches by their own quality and the corresponding classification coupled dictionaries are learned.In the end,single SR fingerprint is reconstructed using sparse representation with ridge pattern by classification coupled dictionaries.