Study On The Low-quality Fingerprint Image Enhancement Based On STFT

With the development of information science and technology,fingerprint recognition technology has been comprehensively promoted,and people have put forward higher requirements on its accuracy,security and usability.In practical applications,due to the insufficient effective information in low-quality fingerprint images,a large number of pseudo minutia are extracted in the stage of feature extraction,which affects the recognition accuracy of low-quality fingerprint images.Researchers usually use fingerprint image enhancement to remove pseudo minutia in low-quality fingerprint images,but traditional enhancement algorithms use the morphology of ridge structure or spectral energy cohesion to estimate filter parameters that are susceptible to noise,which affects the filtering effect and thus reduces the recognition accuracy of low-quality fingerprint images.In this paper,we analyze the low-quality fingerprint images and propose a lowquality fingerprint image enhancement algorithm based on Short Time Fourier Transform(STFT).The orientation and frequency of the low quality fingerprint image remain roughly consistent in the local area with certain local stability.To obtain a good representation of the low quality fingerprint spectrum,STFT is used to chunk the low quality fingerprint image.In addition,the performance of the low-quality fingerprint image enhancement algorithm depends largely on the settings of the filter parameters.In this paper,the low-quality fingerprint image is approximated as an interferometric stripe pattern,and a two-dimensional amplitude modulation frequency modulation signal model is established based on the interferometric stripe pattern.The phase information of the fingerprint determines the texture information of the fingerprint image and represents the intrinsic structural characteristics of the fingerprint.The method in this paper preserves the original information of the fingerprint and improves the filtering effect of low-quality fingerprint images by finding the connection between the phase information and the direction field and frequency field to ensure more accurate parameter estimation of the filter.The experimental results show that the texture information estimated by the algorithm in this paper is less affected by noise,better protects the real stripe structure in the regions with larger stripe curvature of the low-quality fingerprint image,can effectively connect the broken ridges and separate the adhered ridges in the low-quality fingerprint image,increases the number of effective detail points in the low-quality fingerprint image,and then improves the recognition accuracy of the low-quality fingerprint image.