CN113705519B - Fingerprint identification method based on neural network - Google Patents
Fingerprint identification method based on neural network Download PDFInfo
- Publication number
- CN113705519B CN113705519B CN202111034235.9A CN202111034235A CN113705519B CN 113705519 B CN113705519 B CN 113705519B CN 202111034235 A CN202111034235 A CN 202111034235A CN 113705519 B CN113705519 B CN 113705519B
- Authority
- CN
- China
- Prior art keywords
- fingerprint
- images
- image
- features
- fingerprint identification
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Collating Specific Patterns (AREA)
Abstract
The invention relates to the technical field of fingerprint identification, in particular to a fingerprint identification method based on a neural network, which is used for acquiring a preparation image, and a fingerprint sensor with a larger area is used for conventionally adopting 96 x 96 fingerprint images; s2: generating a training image, extracting features by using a fingerprint algorithm, and comparing fingerprint images acquired by the same finger to obtain a partial image A which can be overlapped with each 2 images; and S3, training the images by using arcloss through a full convolution network, wherein the classification accuracy is high, the distance between each type of result is smaller, the distance between different types of result is larger, and the characteristics representing the input 32 x 32 images are trained. And S4, carrying out feature quantization, extracting features, precisely comparing the feature point pairs through coordinates, and meeting the actual corresponding relation of coordinates in the image. The invention solves the problem that on a fingerprint image with a smaller size, because the fingerprint information is carried too little, accurate fingerprint identification is difficult to carry out.
Description
Technical Field
The invention relates to the technical field of fingerprint identification, in particular to a fingerprint identification method based on a neural network.
Background
Fingerprint recognition technology is one of many biometric recognition technologies and is also one of the most widely used technologies. The intelligent mobile phone unlocking, online payment, intelligent door lock, intelligent attendance access control, criminal investigation detect and other occasions are widely applied to machines. Fingerprint recognition must use a fingerprint sensor, and semiconductor fingerprint sensors are most widely used today, and the cost of the semiconductor fingerprint sensor is directly related to the sensor area. With the wide application of fingerprint identification products, the cost of the fingerprint sensor is required to be lower and lower, so that the area of the fingerprint sensor is smaller and smaller. Fingerprint sensor is getting smaller and smaller, can once gather the fingerprint image area also gets smaller and smaller, and the fingerprint information that carries in the image also gets smaller and smaller. The fingerprint identification algorithm which is mature and well used under the condition of large fingerprint image area is difficult to adapt to smaller and smaller fingerprint images. A new fingerprint recognition method must be used. The neural network method is an identification method with wider adaptability and universality obtained by referring to the working principle of the animal central nervous system. The effectiveness of the method has been proved in many occasions, especially the occasion that a large amount of experimental data can be obtained, and the effect of even exceeding the human recognition capability can be achieved.
Scheme one of the prior art: a small-area fingerprint comparison method (patent application number: 201710220456.2) based on deep learning finds feature points in fingerprints, and calculates the direction of the feature points. And rotating and normalizing the image by taking the characteristic point as the center and taking the direction of the characteristic point as the X axis, and cutting out small blocks with set sizes. Training the small blocks by using a depth residual error network to obtain a network for extracting the characteristics. The method must obtain accurate feature points firstly, but the number of feature points in the small-area fingerprint image is small, and the method is difficult to position accurately
Scheme II: fingerprint identification method and terminal equipment (patent application number: 201910064710.3): the refined fingerprint image is obtained through image processing, and then the refined image is input into a network for training, and finally the fingerprint characteristics are obtained. The method needs to obtain a refined fingerprint image. For images with poor quality, it is difficult to obtain accurate detailed images.
Disclosure of Invention
In view of the above, the present invention aims to provide a fingerprint identification method based on a neural network, which is specifically implemented according to the following steps:
s1, acquiring a preparation image, using a fingerprint sensor with a larger area, conventionally adopting 96 x 96 fingerprint images, intensively storing images of the same finger under a folder, adopting at least 2000 images of different fingers, and adopting at least 100 fingerprint images of each finger; the acquired fingerprint images may also include images acquired at longer intervals, with images acquired by different fingerprint sensors.
S2: generating a training image, extracting features by using a fingerprint algorithm, comparing fingerprint images acquired by the same finger to obtain partial images A which can be overlapped with each 2 images, finding partial images A0, A1 and A2 … … with the largest overlapping times in all images of each finger, and cutting the partial images A0, A1 and A2 … … into the size of 32 x 32 to be used as the training image;
S3, training images, namely training out features representing the input 32 x 32 images through arcloss of a full convolution network, wherein the classification accuracy is high, the distance between each type of results is small, and the distance between different types of results is large;
s4, carrying out feature quantization, and carrying out quantization by using a quantization method to obtain quantized features { F0, F1, … … F127 } and finally obtaining 128-dimensional features { F0, F1 … … F127 } which are floating point features; the quantization method is as follows, n=0 is set, 2 different random numbers a, b of 0-127 are generated, fa, fb are compared
If fa > =fb F n=1, otherwise F n=0, n=n+1 up to n > =128, thus obtaining 128 bit features F.
S5, identifying the step1, extracting the characteristics, inputting an image with the size of 64 x 64 to obtain characteristics f of 9 x 128;
S6, recognizing step 2, feature comparison of 2 images A and B to be compared, generating features fA and fB by extracting a feature network, comparing 81 features in fA with 81 features in fB by calculating Hamming distance, and sorting according to the distance to obtain feature point pairs with the minimum m pairs of distances;
S7: and 3, identifying, namely accurately comparing the coordinates with the characteristic point pairs, and meeting the actual corresponding relation of the coordinates in the image.
Further, in step S7, the comparison method is used as follows:
s7.1. first let i=0;
s7.2, extracting feature point pairs in mi, fA (xi, yi), fB (xi, yi);
s7.3, calculating polar coordinates of other points in m by taking fA (xi, yi) as a pole and taking the x direction as a polar axis in the image A;
S7.4:a＝0；
S7.5, rotating the polar coordinate by a degree, converting to an xy coordinate, comparing the xy coordinate with the xy coordinate of the corresponding point in the B, and increasing the matching score if the error is within a certain range;
step S7.6, a=a+1, repeat step S7.5 until a > =360, get the highest matching score Si at a certain value of a;
step S7.7: i=i+1 is repeated until i > =m, yielding the highest matching score Smax at a certain i as the final matching score.
The fingerprint identification method based on the neural network has the beneficial effects that:
1. The direct image is used for training by network input, so that the complex steps of feature points, generating a refined image, and the like, which are required to be extracted in the prior method, are avoided. And training images can be automatically generated, so that manual labeling is avoided.
2. The full convolution network is adopted to train the small image, and the large image is directly used for generating the feature matrix by the full convolution network when the full convolution network is used finally, so that the training process is simple, and the final features are rich.
3. And finally, when the points are compared, the point characteristic comparison and the point coordinates are used in combination, so that a better recognition effect is achieved.
4. The problem that accurate fingerprint identification is difficult to carry out on a fingerprint image with a smaller size due to the fact that the fingerprint information is carried too little is solved.
Drawings
FIG. 1 is a preparation view of an acquired image of the present invention;
FIG. 2 is a diagram of a generated training image of the present invention;
FIG. 3 is a training network diagram of the present invention;
fig. 4 is an identification extraction feature diagram of the present invention.
Detailed Description
The present application will be described in detail below with reference to the drawings and the specific embodiments, and it is apparent that the described embodiments are only some embodiments of the present application, but not all embodiments, and all other embodiments obtained by persons skilled in the art without making creative efforts based on the embodiments in the present application are within the protection scope of the present application.
In this embodiment, the present invention is specifically implemented according to the following steps:
As shown in FIG. 1, S1, collecting preparation images, using a fingerprint sensor with larger area, conventionally adopting 96 x 96 fingerprint images, storing images of the same finger under a folder in a concentrated manner, adopting at least 2000 images of different fingers, and adopting at least 100 fingerprint images of each finger; the acquired fingerprint images may also include images acquired at longer intervals, with images acquired by different fingerprint sensors.
S2: as shown in fig. 2, a training image is generated, features are extracted by using a fingerprint algorithm, fingerprint images acquired by the same finger are compared one by one to obtain partial images a which can be overlapped with each other by 2 images, partial images A0, A1 and A2 … … with the largest overlapping times in all images of each finger are found, and the partial images A0, A1 and A2 … … are cut into the size of 32 x 32 and are used as the training image;
S3, training the image, namely training the features representing the input 32 x 32 image by using arcloss through a full convolution network, wherein the classification accuracy is high, the result distance of each type is small, and the result distance of different types is large;
s4, carrying out feature quantization, and carrying out quantization by using a quantization method to obtain quantized features { F0, F1, … … F127 } and finally obtaining 128-dimensional features { F0, F1 … … F127 } which are floating point features; the quantization method is as follows, n=0 is set, 2 different random numbers a, b of 0-127 are generated, fa, fb are compared
If fa > =fb F n=1, otherwise F n=0, n=n+1 up to n > =128, thus obtaining 128 bit features F.
S5, as shown in FIG. 3, in the step 1, extracting features, and inputting images with the size of 64 x 64 to obtain features f with the size of 9 x 128;
S6, recognizing step 2, feature comparison of 2 images A and B to be compared, generating features fA and fB by extracting a feature network, comparing 81 features in fA with 81 features in fB by calculating Hamming distance, and sorting according to the distance to obtain feature point pairs with the minimum m pairs of distances;
S7: and 3, identifying, namely accurately comparing the coordinates with the characteristic point pairs, and meeting the actual corresponding relation of the coordinates in the image.
In this embodiment, in step S7, the comparison method is used as follows:
s7.1. first let i=0;
s7.2, extracting feature point pairs in mi, fA (xi, yi), fB (xi, yi);
s7.3, calculating polar coordinates of other points in m by taking fA (xi, yi) as a pole and taking the x direction as a polar axis in the image A;
S7.4:a＝0；
S7.5, rotating the polar coordinate by a degree, converting to an xy coordinate, comparing the xy coordinate with the xy coordinate of the corresponding point in the B, and increasing the matching score if the error is within a certain range;
Step S 7.5 is repeated until a > =360, and the highest matching score Si is obtained at a certain value of a;
Step S 7.2 is repeated until i > =m, at which point the highest matching score Smax is obtained as the final matching score.
The above embodiments are only for illustrating the technical solution of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications and equivalents may be made thereto without departing from the spirit and scope of the technical solution of the present invention, which is intended to be covered by the scope of the claims of the present invention. The technology, shape, and construction parts of the present invention, which are not described in detail, are known in the art.
Claims (3)
1. A fingerprint identification method based on a neural network is characterized in that: the method comprises the following steps:
S1, acquiring a preparation image, using a fingerprint sensor, conventionally adopting 96 x 96 fingerprint images, intensively storing images of the same finger under a folder, adopting at least 2000 images of different fingers, and adopting at least 100 fingerprint images of each finger;
s2: generating a training image, extracting features by using a fingerprint algorithm, comparing fingerprint images acquired by the same finger to obtain partial images A which can be overlapped with each 2 images, finding partial images A0, A1 and A2 … … with the largest overlapping times in all images of each finger, and cutting the partial images A0, A1 and A2 … … into the size of 32 x 32 to be used as the training image;
s3, training images, namely training a network model with high classification accuracy, small result distance of each class and large result distance of different classes by using arcloss through a full convolution network;
S4, carrying out feature quantization, and carrying out quantization by using a quantization method to obtain quantized features { F0, F1, … … F127 } and finally obtaining 128-dimensional features { F0, F1 … … F127 }, wherein the 128-dimensional features { F0, F1 … … F127 } are floating point features; the method of quantization is as follows,
Let n=0, generate 2 different random numbers a, b of 0-127, compare Fa, fb,
If fa > = fb, F n = 1, otherwise F n = 0, n = n +1, until n > = 128, thus obtaining 128 bit features F;
S5, identifying the step1, extracting the characteristics, inputting an image with the size of 64 x 64 to obtain characteristics f of 9 x 128;
S6, recognizing step 2, feature comparison of 2 images A and B to be compared, generating features fA and fB by extracting a feature network, comparing 81 features in fA with 81 features in fB by calculating Hamming distance, and sorting according to the distance to obtain feature point pairs with the minimum m pairs of distances;
S7: and 3, identifying, namely accurately comparing the coordinates with the characteristic point pairs, and meeting the actual corresponding relation of the coordinates in the image.
2. The fingerprint identification method based on the neural network according to claim 1, wherein the fingerprint identification method comprises the following steps: in step S7, the comparison method is used as follows:
s 7.1, firstly setting i=0;
S 7.2, extracting a characteristic point pair, fA (xi, yi), fB (xi, yi);
S 7.3, calculating the polar coordinates of the rest points in m by taking fA (xi, yi) as poles and the x direction as polar axes in the image A;
S7.4:a＝0；
S 7.5, rotating the polar coordinate by a degree, converting to an xy coordinate, comparing with the xy coordinate of the corresponding point in the B, and increasing the matching score if the error is in a certain range;
S 7.6 a=a+1, repeating step S 7.5 until a > =360, and obtaining the highest matching score Si at a certain value of a;
Step S 7.2 is repeated until i > =m, at which point i is obtained with the highest matching score Smax as the final matching score, S 7.7 i=i+1.
3. The fingerprint identification method based on the neural network according to claim 1, wherein the fingerprint identification method comprises the following steps: in step S1, the acquired fingerprint image may also contain images acquired by different fingerprint sensors.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111034235.9A CN113705519B (en) | 2021-09-03 | 2021-09-03 | Fingerprint identification method based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111034235.9A CN113705519B (en) | 2021-09-03 | 2021-09-03 | Fingerprint identification method based on neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113705519A CN113705519A (en) | 2021-11-26 |
| CN113705519B true CN113705519B (en) | 2024-05-24 |
Family
ID=78659597
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111034235.9A Active CN113705519B (en) | 2021-09-03 | 2021-09-03 | Fingerprint identification method based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113705519B (en) |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105160303A (en) * | 2015-08-10 | 2015-12-16 | 上海闻泰电子科技有限公司 | Fingerprint identification method based on mixed matching |
| CN107480649A (en) * | 2017-08-24 | 2017-12-15 | 浙江工业大学 | Fingerprint sweat pore extraction method based on full convolution neural network |
| CN108960214A (en) * | 2018-08-17 | 2018-12-07 | 中控智慧科技股份有限公司 | Fingerprint enhancement binarization method, device, equipment, system and storage medium |
| CN109063541A (en) * | 2017-06-08 | 2018-12-21 | 墨奇公司 | System and method for fingerprint recognition |
| CN110427832A (en) * | 2019-07-09 | 2019-11-08 | 华南理工大学 | A kind of small data set finger vein identification method neural network based |
| CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
| CN110991374A (en) * | 2019-12-10 | 2020-04-10 | 电子科技大学 | A fingerprint singularity detection method based on RCNN |
| WO2020083407A1 (en) * | 2018-10-23 | 2020-04-30 | 华南理工大学 | Three-dimensional finger vein feature extraction method and matching method therefor |
| US10664722B1 (en) * | 2016-10-05 | 2020-05-26 | Digimarc Corporation | Image processing arrangements |
| CN111428064A (en) * | 2020-06-11 | 2020-07-17 | 深圳市诺赛特系统有限公司 | Small-area fingerprint image fast indexing method, device, equipment and storage medium |
| CN111429359A (en) * | 2020-06-11 | 2020-07-17 | 深圳市诺赛特系统有限公司 | Small-area fingerprint image splicing method, device, equipment and storage medium |
-
2021
- 2021-09-03 CN CN202111034235.9A patent/CN113705519B/en active Active
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105160303A (en) * | 2015-08-10 | 2015-12-16 | 上海闻泰电子科技有限公司 | Fingerprint identification method based on mixed matching |
| US10664722B1 (en) * | 2016-10-05 | 2020-05-26 | Digimarc Corporation | Image processing arrangements |
| CN109063541A (en) * | 2017-06-08 | 2018-12-21 | 墨奇公司 | System and method for fingerprint recognition |
| CN107480649A (en) * | 2017-08-24 | 2017-12-15 | 浙江工业大学 | Fingerprint sweat pore extraction method based on full convolution neural network |
| CN108960214A (en) * | 2018-08-17 | 2018-12-07 | 中控智慧科技股份有限公司 | Fingerprint enhancement binarization method, device, equipment, system and storage medium |
| WO2020083407A1 (en) * | 2018-10-23 | 2020-04-30 | 华南理工大学 | Three-dimensional finger vein feature extraction method and matching method therefor |
| CN110427832A (en) * | 2019-07-09 | 2019-11-08 | 华南理工大学 | A kind of small data set finger vein identification method neural network based |
| CN110472518A (en) * | 2019-07-24 | 2019-11-19 | 杭州晟元数据安全技术股份有限公司 | A kind of fingerprint image quality judgment method based on full convolutional network |
| CN110991374A (en) * | 2019-12-10 | 2020-04-10 | 电子科技大学 | A fingerprint singularity detection method based on RCNN |
| CN111428064A (en) * | 2020-06-11 | 2020-07-17 | 深圳市诺赛特系统有限公司 | Small-area fingerprint image fast indexing method, device, equipment and storage medium |
| CN111429359A (en) * | 2020-06-11 | 2020-07-17 | 深圳市诺赛特系统有限公司 | Small-area fingerprint image splicing method, device, equipment and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| Presentation attack detection using a tiny fully convolutional network;Eunsoo Park, etal.;《IEEE Transactions on Information Forensics and Security》;第14卷(第11期);全文 * |
| 基于全卷积神经网络的OCT内外指纹提取算法;杨熙丞;《中国优秀硕士学位论文全文数据库》;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113705519A (en) | 2021-11-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jain et al. | On-line fingerprint verification | |
| CN100414558C (en) | Automatic Fingerprint Recognition System and Method Based on Template Learning | |
| Anand et al. | PoreNet: CNN-based pore descriptor for high-resolution fingerprint recognition | |
| CN112597812A (en) | Finger vein identification method and system based on convolutional neural network and SIFT algorithm | |
| CN102254188A (en) | Palmprint recognizing method and device | |
| CN102542243A (en) | LBP (Local Binary Pattern) image and block encoding-based iris feature extracting method | |
| CN110188671B (en) | Method for analyzing handwriting characteristics by using machine learning algorithm | |
| CN1304114A (en) | Identity identification method based on multiple biological characteristics | |
| CN105095880A (en) | LGBP encoding-based finger multi-modal feature fusion method | |
| CN115795394A (en) | Hierarchical Multimodality and Advanced Incremental Learning for Biometric Fusion Identity Recognition | |
| Jain et al. | Fingerprint image analysis: role of orientation patch and ridge structure dictionaries | |
| Nyssen et al. | A multi-stage online signature verification system | |
| Yang et al. | Finger-vein pattern restoration with generative adversarial network | |
| CN113705519B (en) | Fingerprint identification method based on neural network | |
| Mostayed et al. | Biometric authentication from low resolution hand images using radon transform | |
| CN1305001C (en) | Finger print characteristic matching method in intelligent card | |
| Du et al. | Shape matching and recognition base on genetic algorithm and application to plant species identification | |
| Wang et al. | Fingerprint recognition using directional micropattern histograms and LVQ networks | |
| Tahmasebi et al. | Signature identification using dynamic and HMM features and KNN classifier | |
| Arivazhagan et al. | Iris recognition using Ridgelet transform | |
| Anand et al. | Deep convolutional neural network for dot and incipient ridge detection in high-resolution fingerprints | |
| Chowdhury et al. | Efficient fingerprint matching based upon minutiae extraction | |
| Prathiba et al. | Signature verification system based on wavelets | |
| CN111382703A (en) | A Finger Vein Recognition Method Based on Secondary Screening and Score Fusion | |
| Fei et al. | Combining enhanced competitive code with compacted ST for 3D palmprint recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |