A fingerprint image periodic noise removing method, device and computer readable storage medium

Abstract

The present application relates to the fingerprint identification technical field, especially a kind of fingerprint image periodic noise removal method, device and computer readable storage medium, the present application is directed to fixed period, random phase's fingerprint periodic noise, by pre-acquiring noise period and in the frequency domain only acts on fixed frequency point, not with the phase change trap wave filter mask, realize the phase robust noise stable suppression, while accurate keep ridge and minutia, with the advantages of low computational complexity, adapt embedded terminal, improve the identification accuracy of low-quality fingerprint image and robustness.

Description

Technical Field

[0001] This invention relates to the field of fingerprint recognition technology, and in particular to a method, apparatus and computer-readable storage medium for removing periodic noise from fingerprint images. Background Technology

[0002] Fingerprint recognition, due to its uniqueness, stability, and convenience, has been widely used in identity verification, access control, and mobile terminal unlocking. In fingerprint recognition systems, the quality of image preprocessing directly determines the accuracy of subsequent minutiae extraction and matching, with periodic noise removal being particularly crucial. During actual acquisition, fingerprint images are easily affected by factors such as equipment precision (e.g., parallel sampling by multi-channel ADCs) and environmental interference, leading to the introduction of periodic stripe noise. This type of noise has the characteristics of a fixed period and random phase—the period is determined by hardware (e.g., noise introduced by a 16-channel ADC at frequencies multiples of 16), but the phase varies randomly within or between images, severely damaging ridge and valley clarity, resulting in missed or false detections of minutiae and reducing matching accuracy.

[0003] Among existing denoising techniques, spatial filtering (Gaussian and median filtering) is difficult to accurately locate periodic noise and easily blurs ridges; general frequency domain filtering can locate frequency peaks, but it is sensitive to phase, and the notch position is misaligned in random phase, resulting in a sharp drop in denoising effect and possible destruction of high-frequency characteristics; variational methods (TV and BM3D) have good denoising quality but excessive computational complexity, which cannot meet the real-time requirements of embedded terminals, and they also lack targeted design for fixed-period, random-phase noise.

[0004] In summary, existing technologies suffer from problems such as unstable denoising, easy destruction of minutiae, and poor real-time performance. There is an urgent need for a fingerprint periodic noise removal method that can adapt to random phase changes, balance denoising quality and detail preservation, and is computationally efficient. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a method, apparatus, electronic device, and storage medium for removing periodic noise from fingerprint images.

[0006] In a first aspect, embodiments of the present invention provide a method for removing periodic noise from fingerprint images, the method comprising: Obtain the raw fingerprint image; Perform a two-dimensional fast Fourier transform on the original fingerprint image to obtain the frequency domain distribution data; Based on the known periodic noise period of the fingerprint, determine the position of the fixed frequency point corresponding to the periodic noise in the frequency domain; Construct a notch filter mask, which sets a suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase. The frequency domain distributed data is filtered by a notch filter mask to suppress the frequency domain spikes corresponding to periodic noise, and the filtered frequency domain data is obtained. A two-dimensional inverse Fourier transform is performed on the filtered frequency domain data to convert it back to the spatial domain, resulting in a denoised fingerprint image.

[0007] In conjunction with the first aspect, before the step of performing a two-dimensional fast Fourier transform on the original fingerprint image to obtain the frequency domain distribution data, the following steps are also included: Calculate the first mean of each pixel in the original fingerprint image within the first neighborhood and the second mean within the second neighborhood; Calculate the first difference between the pixel grayscale value and the first mean, and the second difference between the pixel grayscale value and the second mean, respectively; The first and second differences are each multiplied by their respective gain coefficients and then superimposed onto the pixel grayscale values ​​of the original fingerprint image to eliminate the influence of baseline inhomogeneity.

[0008] In conjunction with the first aspect, the steps of performing a two-dimensional fast Fourier transform on the original fingerprint image to obtain the frequency domain distribution data include: The original fingerprint image is divided into multiple image blocks; For each image block, a two-dimensional fast Fourier transform is performed on the image block to obtain the frequency domain distribution data corresponding to each image block.

[0009] In conjunction with the first aspect, the step of determining the location of a fixed frequency point corresponding to the periodic noise in the frequency domain based on the known periodic noise period of the fingerprint includes: Obtain the image height of the original fingerprint image; The frequency domain row coordinate interval is obtained by calculating the quotient of the image height and the known fingerprint periodic noise period. Determine the integer multiples of the frequency domain row coordinate intervals, and round each integer multiple to obtain the target row coordinates; Each row corresponding to the target row coordinate in the frequency domain distribution data is determined as a fixed frequency point position.

[0010] In conjunction with the first aspect, the steps of constructing a notch filter mask, wherein the notch filter mask sets a suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase, include: A strip-shaped suppression region with a specified width is set with a fixed frequency point as the center; Set the notch filter mask coefficients within the bar suppression region to the first suppression coefficient; The notch filter mask coefficients of the first part of the bar suppression region that coincides with the main energy region of the fingerprint frequency domain are set as the second suppression coefficients, and the second suppression coefficients are greater than the first suppression coefficients. Set the notch filter mask coefficients of the second part of the main energy region of the fingerprint frequency domain, excluding the first part, as enhancement coefficients, with the enhancement coefficients being greater than 1; A notch filter mask is generated based on the set first suppression coefficient, second suppression coefficient, and enhancement coefficient.

[0011] In conjunction with the first aspect, the steps of using a notch filter mask to filter the frequency domain distributed data, suppressing frequency domain spikes corresponding to periodic noise, and obtaining the filtered frequency domain data include: The notch filter mask is multiplied element-wise with the frequency domain distribution data to obtain the filtered frequency domain data.

[0012] In conjunction with the first aspect, the steps of performing a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain the denoised fingerprint image include: Perform a two-dimensional inverse Fourier transform on the filtered frequency domain data to obtain the complex domain result; Extract the real part of the result in the complex field; Divide the real part by the total number of pixels in the original fingerprint image to obtain the normalized image data; The normalized image data is cropped in grayscale range to limit it to the effective grayscale range; The cropped data is converted into an integer image format to obtain a denoised fingerprint image.

[0013] Following the first aspect, after performing a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain the denoised fingerprint image, the method further includes: Orientation field enhancement is performed on the denoised fingerprint image, smoothing enhancement is performed along the fingerprint ridge direction, and details are preserved in the direction perpendicular to the ridge, to obtain the enhanced image; The enhanced image is binarized to obtain a binary image; The binarized image is thinned, and the processed fingerprint image is output.

[0014] Secondly, this application provides a fingerprint image periodic noise removal apparatus, the apparatus comprising: The acquisition module is used to acquire the raw fingerprint image; The transformation module is used to perform a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data; The determination module is used to determine the position of a fixed frequency point in the frequency domain corresponding to the periodic noise based on the known periodic noise period of the fingerprint. The module is used to construct a notch filter mask. The notch filter mask sets the suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase. The suppression module is used to filter the frequency domain distributed data using a notch filter mask, suppressing the frequency domain spikes corresponding to periodic noise, and obtaining filtered frequency domain data. The conversion module is used to perform a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain the denoised fingerprint image.

[0015] Thirdly, this application provides a computer-readable storage medium storing computer program instructions, which, when read and executed by a processor, perform the above-described method.

[0016] The embodiments of the present invention bring the following beneficial effects: This application provides a fingerprint image periodic noise removal method, apparatus, and computer-readable storage medium. The method includes: acquiring an original fingerprint image; performing a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data; determining the fixed frequency point position corresponding to the periodic noise in the frequency domain based on the known fingerprint periodic noise period; constructing a notch filter mask, wherein the notch filter mask sets a suppression coefficient at the fixed frequency point position, and the fixed frequency point position depends only on the noise period and does not change with the noise phase; using the notch filter mask to filter the frequency domain distribution data to suppress the frequency domain spikes corresponding to the periodic noise, obtaining filtered frequency domain data; performing a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain, obtaining a denoised fingerprint image.

[0017] This invention addresses periodic noise in fingerprints with fixed periods and random phases. By pre-acquiring the noise period and constructing a notch filter mask in the frequency domain that operates only at fixed frequencies and does not change with phase, it achieves phase-robust noise stability suppression while accurately preserving ridges and minutiae. It has the advantages of low computational complexity and compatibility with embedded terminals, thus improving the recognition accuracy and robustness of low-quality fingerprint images.

[0018] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0019] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0020] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 A schematic flowchart of the fingerprint image periodic noise removal method provided in an embodiment of the present invention; Figure 2 A schematic diagram of the fingerprint image periodic noise removal device provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the original fingerprint image provided in an embodiment of the present invention; Figure 4 To Figure 3 The preprocessed base image is obtained by superimposing dual-scale mean differences and limiting the amplitude of the original fingerprint image shown. Figure 5 To Figure 4 A schematic diagram of the frequency domain amplitude distribution obtained after performing a two-dimensional fast Fourier transform; Figure 6 A schematic diagram of a notch filter mask constructed based on a fixed frequency point position; Figure 7 This is a schematic diagram comparing the frequency domain amplitude spectra before and after filtering. Figure 8 This is the denoised fingerprint image; Figure 9 This is the final fingerprint image after orientation field enhancement, binarization, and thinning.

[0022] Figure label: 10 - Acquisition module, 20 - Transformation module, 30 - Determination module, 40 - Construction module, 50 - Suppression module, 60 - Conversion module. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] To facilitate understanding of this embodiment, the technical terms used in this application will be briefly introduced below.

[0025] The Fast Fourier Transform (FFT) is an efficient algorithm for converting signals from the time domain (or spatial domain) to the frequency domain.

[0026] Ridge-valley contrast is the degree of grayscale difference between ridges (dark) and valleys (bright) in a fingerprint image.

[0027] Fixed frequency point position refers to the discrete frequency coordinates in the frequency domain that correspond to fixed periodic noise in the spatial domain. It depends only on the noise period and image size and is not affected by the noise phase.

[0028] After introducing the technical terms used in this application, the application scenarios and design concepts of the embodiments of this application will be briefly described below.

[0029] In fingerprint recognition, the quality of image preprocessing directly affects the recognition accuracy, with periodic noise removal being a key challenge. Existing spatial filtering, general frequency filtering, and variational methods suffer from unstable denoising, loss of minutiae, or poor real-time performance, and there is a particular lack of dedicated solutions for fixed-period, random-phase noise.

[0030] Based on this, embodiments of this application provide a method, apparatus, and computer-readable storage medium for removing periodic noise from fingerprint images.

[0031] Example 1 This application provides a method for removing periodic noise from fingerprint images, combined with... Figure 1 As shown, the method includes: S110, acquire the raw fingerprint image.

[0032] S120 performs a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data.

[0033] S130, based on the known periodic noise period of the fingerprint, determine the position of the fixed frequency point corresponding to the periodic noise in the frequency domain.

[0034] S140, construct a notch filter mask. The notch filter mask sets a suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase.

[0035] S150 uses a notch filter mask to filter the frequency domain distributed data, suppressing the frequency domain spikes corresponding to periodic noise, and obtaining the filtered frequency domain data.

[0036] S160 performs a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain, obtaining the denoised fingerprint image.

[0037] This invention addresses periodic noise in fingerprints with fixed periods and random phases. By pre-acquiring the noise period and constructing a notch filter mask in the frequency domain that operates only at fixed frequencies and does not change with phase, it achieves phase-robust noise stability suppression while accurately preserving ridges and minutiae. It has the advantages of low computational complexity and compatibility with embedded terminals, thus improving the recognition accuracy and robustness of low-quality fingerprint images.

[0038] In step S110, the original fingerprint image is typically output directly by the fingerprint acquisition device, especially suitable for grayscale images acquired by capacitive fingerprint sensors. Combined with... Figure 3 The image is a raw fingerprint grayscale image directly output by the acquisition device. The image contains obvious periodic stripe noise (manifested as alternating bright and dark stripes at fixed intervals). At the same time, due to uneven pressing, the image brightness and contrast vary in different areas, which intuitively reflects the morphology of the periodic noise to be removed.

[0039] Due to factors such as the hardware architecture of the acquisition device (e.g., using 16 ADCs for parallel sampling) and skin condition, contact force, and environmental interference during the pressing process, the original fingerprint image often contains periodic stripe noise. The period of this type of noise is determined by hardware parameters such as the number of sampling channels and the driving clock, and has a fixed spatial frequency, but its phase can change randomly with different acquisition times and different pressing positions. The original fingerprint image contains not only the ridge and valley texture used for identification, but also the aforementioned periodic noise, as well as interference such as uneven illumination and sensor substrate noise.

[0040] In this step, no additional preprocessing or filtering of the image is required; the raw grayscale data output by the sensor can be read directly. The purpose is to preserve the true texture of the fingerprint and the original distribution of noise to the greatest extent possible, providing complete input information for subsequent frequency domain analysis and accurate denoising. The quality and resolution of the original image (such as 64×64, 128×128, or 256×256) will affect the frequency resolution of the subsequent FFT transform, but the fixed-frequency notch filtering method proposed in this invention is adaptive to image size and can work stably as long as the row coordinates corresponding to the noise period are known.

[0041] In conjunction with the first aspect, prior to step S120, the following also includes: S111, calculate the first mean of each pixel in the original fingerprint image within the first neighborhood and the second mean within the second neighborhood.

[0042] S112, calculate the first difference between the pixel grayscale value and the first mean, and the second difference between the pixel grayscale value and the second mean.

[0043] S113, after multiplying the first difference and the second difference by their respective gain coefficients, they are superimposed onto the pixel grayscale values ​​of the original fingerprint image to eliminate the influence of baseline inhomogeneity.

[0044] The raw images acquired by fingerprint sensors often exhibit baseline inhomogeneity, meaning that the background grayscale values ​​vary slowly across different regions within the same image. This can affect the accuracy of subsequent frequency domain analysis and the contrast between ridges and valleys. To eliminate this inhomogeneity, step S111 employs a dual-scale local mean estimation method.

[0045] Specifically, for each pixel in the original fingerprint image, a first neighborhood (e.g., a 7×7 pixel window) and a second neighborhood (e.g., a 13×13 pixel window) are selected centered on that pixel. The arithmetic mean of the grayscale values ​​of all pixels within the window is calculated and denoted as the first mean and the second mean, respectively. The small-scale neighborhood (7×7) can better follow local texture changes but is sensitive to overall brightness shifts; the large-scale neighborhood (13×13) can reflect background brightness trends over a wider range but is not sensitive to detail changes. By using the two means at different scales, the baseline shifts of different frequency components can be separated.

[0046] After obtaining two local means (specifically the first mean and the second mean) in step S111, step S112 further calculates the difference between the original pixel grayscale value and the first mean, which is called the first difference, and the difference between the original pixel grayscale value and the second mean, which is called the second difference.

[0047] The difference reflects the degree of deviation of a pixel value relative to the local background: if the pixel is a ridge (usually darker), the difference may be negative; if it is a valley (brighter), the difference is positive. Since baseline unevenness manifests as slow grayscale changes, this change is approximately constant within the local window. Therefore, the difference between a pixel and the mean mainly contains high-frequency fingerprint texture information and noise, while low-frequency background drift is partially canceled out. The two differences respectively reflect the local contrast of the pixel against different scale backgrounds.

[0048] Step S113 enhances fingerprint texture through weighted superposition while suppressing baseline drift. In this embodiment, the first difference is multiplied by a gain factor of 2, the second difference is multiplied by a gain factor of 2.5, and then the two products are added together to obtain the total difference value. To further enhance contrast, this total difference value is multiplied by 30 times and its value range is limited to the grayscale variation range of [-30, 30] to prevent over-enhancement or saturation artifacts.

[0049] Finally, the processed difference values ​​are superimposed onto the original pixel grayscale values ​​to form the output pixel. This operation is essentially an adaptive high-pass filter: small-scale differences (7×7) are amplified by a factor of 2, and large-scale differences (13×13) are amplified by a factor of 2.5, which enhances local high-frequency components, while large-scale baseline inhomogeneities, which were removed in the difference calculation, are not recovered after superposition. By limiting the difference range, grayscale overflow caused by excessive noise is avoided.

[0050] Images processed by S111 to S113 exhibit enhanced contrast between ridges and valleys, with more uniform background brightness, providing a cleaner and more stable data source. Figure 4 To Figure 3 The original fingerprint image shown is the preprocessed base image obtained after superimposing the mean difference between 7×7 and 13×13 scales and then limiting the amplitude. Figure 3 In comparison, this Figure 4 The baseline inhomogeneity was effectively eliminated, and the local contrast between ridges and valleys was improved, but periodic stripe noise still existed.

[0051] In conjunction with the first aspect, step S120 includes: S121, the original fingerprint image is divided into multiple image blocks.

[0052] S122, For each image block, perform a two-dimensional fast Fourier transform on the image block to obtain the frequency domain distribution data corresponding to each image block.

[0053] After dividing the image into blocks, this step performs a two-dimensional Fast Fourier Transform (FFT) on each block. Since each block is relatively small (e.g., 64×64), the FFT computation is significantly less than that of the entire image, improving overall computational efficiency. After the transform, the frequency domain distribution data for each block is a complex matrix, where the number of rows and columns equals the height and width of the block, respectively. Each complex element contains the amplitude and phase information of a spatial frequency component within that block. For fixed-period stripe noise, it will also appear as an energy spike at a fixed row coordinate in the frequency domain of each block. This row coordinate depends only on the block height and the known noise period, and is independent of the block content. Therefore, subsequent fixed-frequency point determination, notch mask construction, and filtering operations can be performed independently on the frequency domain data of each block. Because the block sizes are identical, the formulas for calculating the frequency positions corresponding to the noise period are completely consistent, and the mask structure can be reused, greatly reducing implementation complexity. After all image blocks have been processed, the denoised sub-images obtained by inverse transformation of each image block are stitched together according to the original division order to restore the complete denoised fingerprint image.

[0054] Figure 5 To Figure 4 The frequency domain amplitude distribution obtained after performing a two-dimensional fast Fourier transform (2D-FFT) (usually logarithmically stretched for observation). Figure 5 The periodic bright spots located on fixed row coordinates can be clearly seen, which are the frequency domain spikes corresponding to periodic noise. These spikes are arranged at equal intervals along the vertical direction, corresponding to stripe noise with a spatial domain period of 8 pixels.

[0055] In conjunction with the first aspect, step S130 includes: S131, Obtain the image height of the original fingerprint image.

[0056] S132, calculate the quotient of the image height and the known fingerprint periodic noise period to obtain the frequency domain row coordinate interval.

[0057] S133, determine the integer multiples of the frequency domain row coordinate interval, and round each integer multiple to obtain the target row coordinates.

[0058] S134 determines the fixed frequency point positions of each row corresponding to the target row coordinates in the frequency domain distribution data.

[0059] In the two-dimensional discrete Fourier transform, the row coordinates of the frequency domain matrix (usually corresponding to the spatial frequency in the vertical direction) are directly related to the height of the original image. Specifically, if the height of the original fingerprint image is H (number of pixels), then after the 2D-FFT transform, the number of rows of the frequency domain distribution data is also H, where the vertical spatial frequency corresponding to the u-th row (u=0, 1…H-1) is (period u / pixel H). To map the known spatial domain noise period (e.g., T=8 pixels) to the row coordinates in the frequency domain, the image height H must first be obtained.

[0060] Step S131 directly reads the value from the size information of the original fingerprint image as the basis for subsequent calculations. It should be noted that if the image has been segmented before FFT, then H should be the height of the segmented image blocks, depending on the implementation method. In this invention, whole-image transformation or fixed segmentation is typically used to ensure consistency in frequency resolution.

[0061] Step S132: It is known that the fingerprint periodic noise has a fixed repetition period T (in pixels / period) in the spatial domain. According to the scaling characteristics of the Fourier transform, the fringe noise with a period of T in the spatial domain is represented by a series of discrete peaks in the frequency domain. The row coordinate interval of these peaks in the vertical direction is: Δv=H / T.

[0062] For example, with T=8 pixels and image height H=1500 pixels, Δv=187.5 is calculated. Since frequency domain coordinates must be integers, in actual processing, this value will be rounded or a bar suppression region will be used to cover multiple nearby rows.

[0063] In this embodiment, the noise period T is known to be 8 pixels, and this period is determined by the hardware parameters of the acquisition device. For example, when the acquisition instrument uses a 16-channel ADC for parallel sampling, its sampling clock or channel crosstalk will generate periodic stripes with a spatial frequency that is a multiple of 16, which are represented as noise with a period of approximately 8 pixels in the image domain (the actual period depends on the ratio of sensor resolution to sampling rate). This fixed period can be accurately determined by pre-analyzing the output image of the acquisition device.

[0064] Step S132 obtains the interval value through division, which determines the repetition period of the noise spike in the frequency domain. Since the noise period T is a fixed value determined by the hardware of the acquisition device (such as a 16-channel ADC), Δv is also fixed and does not change with the image content or noise phase. This is the basis for the phase robust filtering of this invention.

[0065] In practical applications, the energy spikes corresponding to periodic noise in the frequency domain do not appear on only one row coordinate, but on multiple row coordinates with equal intervals of Δv, i.e., Δv, 2Δv, 3Δv… up to no more than the image height H. Furthermore, symmetrical positions near zero frequency are usually included (if the noise is a real signal, the frequency domain is conjugate symmetrical). Step S133 calculates a series of target row coordinate values ​​by calculating integer multiples of this interval (including 1x, 2x, etc.). For example, if Δv = 187.5, the target row coordinates could be 187.5, 375, 562.5, etc. In actual implementation, a rounding strategy (rounding down, rounding up, or taking the nearest integer interval) needs to be combined. These integer multiple coordinates correspond to the fundamental frequency and the positions of each harmonic in the frequency domain. Since the energy of periodic noise is mainly concentrated on the fundamental frequency and its lower harmonics, it is usually sufficient to suppress the rows corresponding to the first few integer multiple coordinates to effectively remove the noise while avoiding over-filtering.

[0066] After determining the target row coordinates in step S133, step S134 marks the entire row element (or a column within a specified range within the row) pointed to by these row coordinates in the frequency domain distribution data as fixed frequency point positions. These are called fixed frequency point positions because they depend only on the noise period T and the image height H, and are independent of the specific phase value of the noise. Regardless of whether the starting phase of the periodic stripes in the current fingerprint image is 0 degrees, 90 degrees, or any other arbitrary angle, its frequency domain energy peaks will appear at these same row coordinates (phase information only affects the distribution of the real and imaginary parts in the complex domain, and does not change the frequency coordinates of the energy concentration). Therefore, the frequency point positions determined in step S134 have phase robustness. Subsequent notch filtering masks will suppress these fixed positions, thereby achieving a stable and consistent denoising effect.

[0067] In practical implementation, since Δv may be non-integer, a strip region with a width of r (e.g., 1 to 3 pixel rows) can be used to cover the vicinity of the target row coordinates to ensure that the noise energy is fully contained.

[0068] In conjunction with the first aspect, step S140 includes: S141, with a fixed frequency point as the center, a stripe suppression region with a specified width is set.

[0069] S142, set the notch filter mask coefficients within the bar suppression region to the first suppression coefficient.

[0070] S143, set the notch filter mask coefficient of the first part of the bar suppression region that coincides with the main energy region of the fingerprint frequency domain as the second suppression coefficient, the second suppression coefficient being greater than the first suppression coefficient.

[0071] S144, set the notch filter mask coefficients of the second part (excluding the first part) within the main energy region of the fingerprint frequency domain to the enhancement coefficients, with the enhancement coefficients being greater than 1.

[0072] S145 generates a notch filter mask based on the set first suppression coefficient, second suppression coefficient, and enhancement coefficient.

[0073] In frequency domain distributed data, the energy of periodic noise is not strictly concentrated on a single row coordinate, but is distributed within a specified range near a fixed frequency point. This is mainly due to spectral leakage caused by the limited image size, and the noise itself may have small periodic fluctuations.

[0074] To effectively capture and suppress noise energy, step S141 sets a bar-shaped suppression region with a specified width along the row direction (vertical frequency axis) centered on the fixed frequency point position (i.e., the target row coordinates) determined in step S130. The width of the bar-shaped suppression region can be set according to the actual noise bandwidth, for example, 1 to 3 pixels. This bar-shaped suppression region extends along the column direction (horizontal frequency axis), covering all or most of the columns, because periodic noise usually appears as horizontal or vertical stripes in the entire image, and its frequency domain energy distribution is a straight line passing through the origin. By setting the bar-shaped region, the frequency band where noise energy is concentrated can be fully covered, while avoiding unnecessary suppression of non-noise frequencies.

[0075] In this embodiment, the fingerprint frequency domain main energy region refers to the region in the frequency domain where the main frequency components corresponding to the fingerprint ridge texture are concentrated. It is usually represented as a low-frequency to mid-frequency circular / rectangular region centered at zero frequency. It can be determined by statistically analyzing the frequency domain amplitude spectrum of a large number of fingerprint samples. For example, a circular region with a radius of 1 / 3 of the image height can be taken, or it can be directly set as the center region of the frequency domain (such as a rectangle with row coordinates in [-H / 4, H / 4] and column coordinates in [-W / 4, W / 4]).

[0076] To suppress periodic noise, the amplitude of the frequency domain components within the bar suppression region needs to be reduced. Step S142 sets the values ​​of all elements in the mask corresponding to the bar suppression region to the first suppression coefficient. The first suppression coefficient is a positive number less than 1; for example, 0.2 indicates that the frequency domain energy of the bar suppression region is attenuated to 20% of its original value. Since this bar suppression region is precisely where the noise peaks are located, multiplying it by a coefficient less than 1 weakens the noise energy, thereby removing periodic stripes in the spatial domain.

[0077] In practical applications, the specific value of the first suppression coefficient can be adjusted according to the noise intensity and image quality. The smaller the value, the stronger the noise reduction effect, but it may cause slight damage to fingerprint textures with similar angular frequencies.

[0078] Not all frequency components in a fingerprint image contain useful information in the frequency domain. The low-frequency region (corresponding to the slowly varying ridge direction) and the mid-to-high-frequency region (corresponding to ridge edges, minutiae, and other details) are the main energy regions. If the bar suppression region happens to overlap with the main energy region of the fingerprint's frequency domain, simply using the first suppression coefficient will excessively attenuate the fingerprint's effective features, leading to loss of detail. Therefore, step S143 performs special processing on this overlapping region: the mask coefficient is set to a second suppression coefficient, which is greater than the first suppression coefficient (e.g., 0.7), meaning that only mild suppression, rather than complete attenuation, is applied to this region. This removes noise components superimposed on the fingerprint texture while retaining most of the fingerprint's inherent frequency domain information, achieving a balance between noise reduction and detail preservation.

[0079] The portion of the fingerprint's main energy region that lacks periodic noise energy (i.e., the region outside the stripe suppression region) contains the crucial texture information required for fingerprint recognition. To further improve image quality and enhance ridge-valley contrast, step S144 sets the masking coefficients of these regions to enhancement coefficients, which are greater than 1 (e.g., 1.3 in a preferred implementation). By multiplying by a coefficient greater than 1, the frequency domain components of this region can be amplified, thereby enhancing the clarity of the fingerprint ridges in the spatial domain.

[0080] It is important to note that the enhancement coefficient should not be too large, otherwise it may introduce new artifacts or amplify residual noise. It is usually determined by combining experience or experiments to achieve the best accuracy in extracting details from the enhanced image.

[0081] After setting the coefficients for each of the above-mentioned bar suppression regions, step S145 generates a mask matrix with the same size as the frequency domain distributed data. Each element is assigned a corresponding coefficient value according to its region: the part of the bar suppression region that does not overlap with the fingerprint main energy is the first suppression coefficient; the part of the bar suppression region that overlaps with the fingerprint main energy is the second suppression coefficient; the part of the fingerprint main energy region other than the bar suppression region is the enhancement coefficient; and the remaining uninvolved regions (such as high-frequency noise regions and non-fingerprint main energy regions) can be kept as 1 (no processing). The final notch filter mask is a two-dimensional matrix that has both noise suppression and fingerprint enhancement functions. Since the fixed frequency point position depends only on the noise period, the mask structure is not sensitive to noise phase changes, ensuring the stability and robustness of the filtering effect.

[0082] Combination Figure 6As shown, in the constructed notch filter mask, the dark bar areas (rows with a width of 1-3 pixels) correspond to the first suppression coefficient (e.g., 0.2), the light gray areas correspond to the second suppression coefficient (e.g., 0.7), and the white areas correspond to the enhancement coefficient (e.g., 1.3) or the retention coefficient 1. This mask only applies to specific row coordinates and does not change with the noise phase, thus achieving phase robustness.

[0083] In conjunction with the first aspect, step S150 includes: S151, multiply the notch filter mask element-by-element with the frequency domain distribution data to obtain the filtered frequency domain data.

[0084] After constructing the notch filter mask, step S150 applies it to the frequency domain distributed data to achieve precise suppression of periodic noise. Specifically: The frequency domain distributed data (a two-dimensional complex matrix) is multiplied element-wise by a notch filter mask (a real matrix of the same size as the frequency domain distributed data). For each frequency element (i.e., each complex point) in the frequency domain distributed data, its amplitude is multiplied by the coefficient at the corresponding position in the mask, while the phase remains unchanged. The magnitude of the mask coefficients determines the degree to which the frequency component is preserved, suppressed, or enhanced.

[0085] Because the notch filter mask has a suppression coefficient less than 1 at a fixed frequency point (and its bar suppression region), the complex amplitude of the frequency domain corresponding to the noise spikes is reduced after the multiplication operation. This significantly weakens or even eliminates the intensity of the periodic stripes during the subsequent inverse transformation back to the spatial domain. Simultaneously, for the portions of the fingerprint's main energy region that do not overlap with noise, the mask coefficient is an enhancement coefficient greater than 1. The multiplication operation amplifies the amplitude of these frequency components, thereby improving the contrast between ridges and valleys in the spatial domain. For regions with a mask coefficient of 1, the frequency domain data remains unchanged, neither introducing additional noise nor damaging the original fingerprint texture.

[0086] As is understandable, element-wise multiplication is a linear filtering operation, characterized by high computational efficiency and ease of hardware acceleration. Compared to complex adaptive filtering or iterative optimization, this operation only requires a single multiplication to traverse all frequency domain units, effectively controlling computational complexity and meeting the real-time requirements of embedded terminals. The resulting filtered frequency domain data removes frequency domain spikes from periodic noise while retaining the effective frequency domain characteristics of the fingerprint. All operations are performed in the complex domain, without introducing phase distortion.

[0087] Combination Figure 7 As shown, Figure 7 The image obtained after applying a notch mask. Figure 7 The noise spikes shown in the image have been suppressed, while the amplitude of the main energy region of the fingerprint has been appropriately enhanced.

[0088] In conjunction with the first aspect, step S160 includes: S161 performs a two-dimensional inverse Fourier transform on the filtered frequency domain data to obtain the complex domain result.

[0089] S162, extract the real part of the result in the complex field.

[0090] S163, divide the real part by the total number of pixels in the original fingerprint image to obtain normalized image data.

[0091] S164 performs grayscale range cropping on the normalized image data to limit it to the effective grayscale range.

[0092] S165 converts the cropped data into an integer image format to obtain a denoised fingerprint image.

[0093] The filtered frequency domain data in step S150 is a two-dimensional complex matrix, where each complex number contains the amplitude and phase information of that frequency component. To recover the spatial domain image from the frequency domain, step S161 performs a two-dimensional discrete inverse Fourier transform (2D-IFFT), re-superimposing the frequency components in the frequency domain according to their amplitude and phase to generate the pixel arrangement in the spatial domain. Since the original fingerprint image is a real matrix before the Fourier transform, the transformed frequency domain data has conjugate symmetry; similarly, performing an inverse transform on the filtered frequency domain data, which still maintains conjugate symmetry, should theoretically yield a real spatial domain matrix.

[0094] However, due to floating-point precision errors in numerical calculations and the slight disruption of ideal symmetry caused by filtering operations, the inverse transform result is typically a complex matrix, with its real part corresponding to the actual pixel grayscale value and its imaginary part representing a tiny residual close to zero. This complex-domain result is the direct output of the IFFT.

[0095] In the complex matrix output by the two-dimensional inverse Fourier transform, each element contains both a real and an imaginary part. Theoretically, after symmetry-preserving filtering, the imaginary part should be approximately zero. However, due to the finite word length effect and the non-ideal symmetry of the filter mask, a small imaginary part value is generated in actual calculations. These imaginary parts are not effective image information. Directly taking the complex modulus or ignoring the real part may introduce phase errors or reduce image accuracy.

[0096] Therefore, step S162 explicitly extracts the real part of each element in the complex matrix and discards the imaginary part. The operation of extracting the real part results in a real matrix with the same size as the original image. Theoretically, its numerical range should be close to the original grayscale range, but after filtering and scaling of the enhancement coefficients, the numerical range may change.

[0097] In commonly used FFT library implementations, the forward two-dimensional Fourier transform (FFT) typically does not perform normalization, while the inverse IFFT implementation may or may not include a normalization factor of 1 / (H×W), depending on the algorithm conventions. To ensure correct grayscale values ​​regardless of the FFT library used, this invention explicitly performs a normalization operation: each element in the extracted real part matrix is ​​divided by the total number of pixels in the original fingerprint image (i.e., image height multiplied by width). After this division, the numerical range is restored to the order of the original grayscale values ​​(e.g., around 0~255). This step ensures that the denoised image has a consistent grayscale scale with the original image.

[0098] Despite the above processing, due to filtering coefficients (such as enhancement coefficient 1.3) and rounding, the normalized image data may contain pixel values ​​exceeding the standard grayscale range [minimum, maximum], such as negative values ​​or values ​​exceeding the maximum grayscale value (such as 255). These outliers are physically meaningless, and directly retaining them would lead to display abnormalities or failure of subsequent processing algorithms. Step S164 performs pixel-by-pixel cropping on the normalized image data: all pixels smaller than the minimum effective grayscale value are set to the minimum value, and all pixels larger than the maximum effective grayscale value are set to the maximum value. The effective grayscale range is usually taken as [0, L], where L is the upper limit of an 8-bit grayscale image (such as 255). In this embodiment, no specific value is limited; it is only used as an example. This cropping ensures the validity of the image data while avoiding the overall contrast of the fingerprint texture due to local overexposure or underexposure.

[0099] The cropped image data is currently in floating-point or double-precision real number format, while most image storage and processing algorithms (such as orientation field filtering, binarization, and minutiae extraction) typically require input in integer pixel format (such as 8-bit unsigned integers). Step S165 converts the cropped data to a standard integer image format, for example, by rounding (rounding to the nearest integer or rounding down), and then converts the data type to an 8-bit unsigned integer, with each pixel occupying one byte and a value range of 0~255. The converted image is the final denoised fingerprint image, whose noise has been suppressed and fingerprint texture has been preserved or enhanced, and can be directly output to the subsequent orientation field enhancement and minutiae extraction modules. At this point, the entire frequency domain filtering process from the original image input to the generation of the denoised image is complete.

[0100] Figure 8 The image shows a spatial domain fingerprint image after IFFT reconstruction, grayscale cropping, and type conversion. Periodic stripe noise has been largely eliminated, ridges and valleys are clear, details are well preserved, and there is no ringing effect or blurring of details caused by filtering.

[0101] In conjunction with the first aspect, after step S160, the following also includes: S170 performs orientation field enhancement on the denoised fingerprint image, performs smooth enhancement along the fingerprint ridge direction, and preserves details in the direction perpendicular to the ridge, thus obtaining the enhanced image.

[0102] S180, perform binarization processing on the enhanced image to obtain a binarized image.

[0103] S190 performs thinning processing on the binarized image and outputs the processed fingerprint image.

[0104] After frequency domain filtering and noise reduction, the periodic stripes in the fingerprint image have been effectively suppressed, but the contrast between ridges and valleys may still be insufficient, and the edge sharpness of local areas needs to be improved. To further highlight the fingerprint texture structure and facilitate subsequent minutiae extraction, step S170 employs directional field enhancement technology.

[0105] The local ridge orientation (i.e., the tangent of the fingerprint pattern) at each pixel in the image is estimated, which can be obtained through gradient analysis, directional filter banks, and other methods. Then, smoothing enhancement is performed along the ridge orientation. This involves using anisotropic filters (such as directional Gaussian filtering or Gabor filtering) to perform low-pass filtering in the direction parallel to the ridges to connect broken ridges and fill in micro-holes; simultaneously, high-frequency components are preserved or enhanced in the direction perpendicular to the ridges to sharpen the boundaries between ridges and valleys. This orientation-adaptive filtering method improves the clarity of the fingerprint texture without blurring the location of details. The output is an enhanced grayscale image with stronger ridge-valley contrast and better texture continuity.

[0106] To convert the grayscale image into a binary form (ridges are black, valleys are white, or vice versa) that facilitates minutiae detection, binarization is further performed in step S180. Considering the potential differences in brightness and contrast across different regions of the fingerprint image, using a globally fixed threshold often yields poor results. Therefore, an adaptive thresholding method is used: for each pixel, a threshold is dynamically calculated based on the mean grayscale value or Gaussian weighted mean of its local neighborhood. If the pixel's grayscale value is greater than the threshold, it is classified as a valley (white); otherwise, it is classified as a ridge (black). Adaptive thresholding effectively overcomes problems such as uneven illumination and uneven pressure, and is particularly suitable for images after directional enhancement, ensuring uniform ridge width and good connectivity in the binarized result. The output is a binary image, with each pixel containing only two states, clearly indicating the skeleton region of the fingerprint ridges.

[0107] Ridges in a binarized image have a specified width, which increases the complexity and false alarm rate of minutiae detection (endpoints, bifurcation points). Thinning (also known as skeletonization) peels away the ridges layer by layer from the binarized image until each ridge shrinks to a single-pixel-wide centerline (skeleton), while preserving the original topological structure (such as connectivity, endpoint, and bifurcation point positions). Common thinning algorithms include morphological thinning and fast lookup table thinning. After thinning, the fingerprint image becomes a single-pixel-wide line graph, from which minutiae such as endpoints and bifurcation points can be accurately and efficiently extracted. The output processed fingerprint image is this thinned image, which can be directly input into the minutiae extraction module of the fingerprint recognition algorithm, providing high-quality input for subsequent feature matching.

[0108] Figure 9 To Figure 8 The image shown is the final image obtained after the denoised image is subjected to directional Gaussian filtering (or Gabor filtering), binarization, and thinning. Figure 9 The mid-ridge line has been smoothed into a single-pixel-width skeleton, with clear details such as endpoints and bifurcation points, which can be directly used for fingerprint feature extraction and matching.

[0109] Secondly, embodiments of this application also provide a fingerprint image periodic noise removal device, combined with Figure 2 As shown, the device includes: an acquisition module 10, a transformation module 20, a determination module 30, a construction module 40, a suppression module 50, and a conversion module 60.

[0110] The acquisition module 10 is used to acquire the original fingerprint image.

[0111] The transformation module 20 is used to perform a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data.

[0112] The determination module 30 is used to determine the position of a fixed frequency point in the frequency domain corresponding to the periodic noise based on the known periodic noise period of the fingerprint.

[0113] The construction module 40 is used to construct a notch filter mask. The notch filter mask sets a suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase.

[0114] The suppression module 50 is used to filter the frequency domain distributed data using a notch filter mask, suppress the frequency domain spikes corresponding to periodic noise, and obtain filtered frequency domain data.

[0115] The conversion module 60 is used to perform a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain a denoised fingerprint image.

[0116] Thirdly, embodiments of this application provide a computer-readable storage medium storing computer program instructions, which are read and executed by a processor to perform the above-described method.

[0117] The storage media described in this application include, but are not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. These storage media can be adapted to embedded fingerprint recognition terminals (such as access controllers, smart locks, and mobile terminal fingerprint modules) or servers or general-purpose computing devices. When the instructions in the storage media are loaded into the device's processor and executed, the fingerprint image periodic noise removal method provided in this application can be implemented, completing all steps from acquiring the original fingerprint image to outputting the denoised image.

[0118] Because the method provided in this application has low computational complexity and requires no iterative optimization, it can meet real-time processing requirements even when running on resource-constrained embedded terminals. Any electronic device with data processing capabilities, as long as it can read the instructions in the storage medium, can perform periodic noise removal on fingerprint images based on the technical solution of this application, thereby improving the overall performance of the fingerprint recognition system.

[0119] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system and apparatus described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0120] Furthermore, in the description of the embodiments of the present invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention based on the specific circumstances.

[0121] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0122] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0123] Finally, it should be noted that the above embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for removing periodic noise from fingerprint images, characterized in that, The method includes: Obtain the raw fingerprint image; Perform a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data; Based on the known periodic noise period of the fingerprint, determine the position of the fixed frequency point in the frequency domain corresponding to the periodic noise; Construct a notch filter mask, wherein the notch filter mask has a suppression coefficient at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase; The frequency domain distribution data is filtered using the notch filter mask to suppress the frequency domain spikes corresponding to periodic noise, resulting in filtered frequency domain data. A two-dimensional inverse Fourier transform is performed on the filtered frequency domain data to convert it back to the spatial domain, resulting in a denoised fingerprint image.

2. The method according to claim 1, characterized in that, Before the step of performing a two-dimensional fast Fourier transform on the original fingerprint image to obtain the frequency domain distribution data, the method further includes: Calculate the first mean of each pixel in the original fingerprint image within the first neighborhood and the second mean within the second neighborhood; Calculate the first difference between the pixel grayscale value and the first mean, and the second difference between the pixel grayscale value and the second mean, respectively; The first difference and the second difference are each multiplied by their corresponding gain coefficients and then superimposed onto the pixel grayscale values ​​of the original fingerprint image to eliminate the influence of baseline inhomogeneity.

3. The method according to claim 1, characterized in that, The step of performing a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data includes: The original fingerprint image is divided into multiple image blocks; For each image block, a two-dimensional fast Fourier transform is performed on the image block to obtain the frequency domain distribution data corresponding to each image block.

4. The method according to claim 1, characterized in that, The step of determining the position of a fixed frequency point in the frequency domain corresponding to the known periodic noise of a fingerprint, based on the known periodic noise period, includes: Obtain the image height of the original fingerprint image; The frequency domain row coordinate interval is obtained by calculating the quotient of the image height and the known periodic noise period of the fingerprint; Determine the integer multiples of the frequency domain row coordinate interval, and round each integer multiple to obtain the target row coordinates; Each row corresponding to the target row coordinate in the frequency domain distribution data is determined as the fixed frequency point position.

5. The method according to claim 1, characterized in that, The step of constructing a notch filter mask, wherein the notch filter mask has a suppression coefficient set at a fixed frequency point, and the fixed frequency point depends only on the noise period and does not change with the noise phase, includes: A strip-shaped suppression region with a specified width is set with the fixed frequency point as the center; Set the notch filter mask coefficients within the strip suppression region to the first suppression coefficient; The notch filter mask coefficient of the first part of the strip suppression region that coincides with the main energy region of the fingerprint frequency domain is set as the second suppression coefficient, and the second suppression coefficient is greater than the first suppression coefficient. The notch filter mask coefficients of the second part of the main energy region of the fingerprint frequency domain, excluding the first part, are set as enhancement coefficients, and the enhancement coefficients are greater than 1. The notch filter mask is generated based on the set first suppression coefficient, second suppression coefficient, and enhancement coefficient.

6. The method according to claim 1, characterized in that, The steps of filtering the frequency domain distribution data using the notch filter mask to suppress frequency domain spikes corresponding to periodic noise and obtaining filtered frequency domain data include: The notch filter mask is multiplied element-wise with the frequency domain distribution data to obtain the filtered frequency domain data.

7. The method according to claim 1, characterized in that, The steps of performing a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain the denoised fingerprint image include: Perform a two-dimensional inverse Fourier transform on the filtered frequency domain data to obtain the complex domain result; Extract the real part of the complex field result; Divide the real part by the total number of pixels in the original fingerprint image to obtain normalized image data; The normalized image data is cropped in grayscale range to limit it to the effective grayscale range; The cropped data is converted into an integer image format to obtain the denoised fingerprint image.

8. The method according to claim 1, characterized in that, After performing a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain the denoised fingerprint image, the method further includes: The denoised fingerprint image is enhanced by directional field enhancement, smoothing enhancement is performed along the fingerprint ridge direction, and details are preserved in the direction perpendicular to the ridge, resulting in an enhanced image; The enhanced image is binarized to obtain a binarized image; The binarized image is thinned to output the processed fingerprint image.

9. A fingerprint image periodic noise removal device, characterized in that, The device includes: The acquisition module is used to acquire the raw fingerprint image; The transformation module is used to perform a two-dimensional fast Fourier transform on the original fingerprint image to obtain frequency domain distribution data; The determination module is used to determine the position of a fixed frequency point in the frequency domain corresponding to the periodic noise based on the known periodic noise period of the fingerprint. A construction module is used to construct a notch filter mask, wherein the notch filter mask has a suppression coefficient set at a fixed frequency point position, and the fixed frequency point position depends only on the noise period and does not change with the noise phase. The suppression module is used to filter the frequency domain distribution data using the notch filter mask, suppressing the frequency domain spikes corresponding to periodic noise, and obtaining filtered frequency domain data. The conversion module is used to perform a two-dimensional inverse Fourier transform on the filtered frequency domain data to convert it back to the spatial domain and obtain a denoised fingerprint image.

10. A computer-readable storage medium, characterized in that, The storage medium stores computer program instructions, which, when read and executed by a processor, perform the method described in any one of claims 1 to 8.

Images