High-end Hardware Manufacturer of POS Terminal & Self-service Kiosk 


new center banner
Home  > INFO CENTER  >  Let you understand fingerprint recognition

Let you understand fingerprint recognition


First, fingerprint recognition into a smart phone standard

With the popularity of mobile payment services, fingerprint recognition technology has become the standard of today's smart phones, and the new technology such as CMOS image sensor / TFT display, ultrasonic detection, etc., has ushered in the development of its market. New Year.

According to research firm Yole, the compound annual growth rate (CAGR) of the fingerprint identification market will reach 19% in the next five years, and the market size is expected to increase from 2.8 billion US dollars in 2016 to 4.7 billion US dollars in 2022.

Initially, it was only used as a component to facilitate the unlocking function of mobile phones. Fingerprint recognition sensors, now driven by the mobile payment service of smart phones, have become an important security element for mobile payment. According to industry analysts, most of the current fingerprint identification market comes from the OEM's demand for all-glass design and waterproof function. This has prompted new technologies such as CMOS/TFT and ultrasonic detection to further promote the evolution of highly integrated fingerprint recognition technology.

According to statistics, the shipment of fingerprint recognition sensors in 2016 has reached 689 million, compared with 23 million in 2013, CAGR reached 210%. Of course, a large number of demand has also led to a lower average price of fingerprint recognition sensors, which has now fallen from $5 to $3, or even lower, and suppliers will continue to face price pressure in the future.

The fingerprint identification market is highly flexible for sensor manufacturers. Although the market scale is very impressive in the next five years, how to stand out in the fierce market competition is still a difficult problem for the relevant manufacturers.

Fingerprint recognition technology includes the following two main identification technologies:

The first method is to use statistical comparison of different fingerprint images; the second method is to use the characteristic information inherent in the fingerprint image to compare.

The first method mainly compares the two fingerprint images statistically to see the similarity between them, and judges whether the two fingerprints are taken from the same person according to the size, thereby realizing the identity recognition function. The second method is to compare their feature information and confirm their identity based on the structural features of the two fingerprint images. Features contain two types: global feature types and local feature types.

The whole process of fingerprint identification technology is:

(1) Using a fingerprint collection device to collect fingerprint images. (2) Preprocessing a large number of noise points in the fingerprint image, thereby improving the efficiency of subsequent processing. After pre-processing, an outline of the fingerprint image is obtained to prepare for the next feature extraction. (3) Perform feature extraction of the fingerprint image and extract its feature information points. (4) Feature matching is performed on the fingerprint image, and the extracted feature points are compared with the feature points pre-stored in the database, and the identity is determined by comparison. According to a study by British scholar E.R.Herry, in the two fingerprint images, if the logarithm of the feature points is 13 pairs, it can be considered that the two images are taken from the same person.

The main performance parameters of the fingerprint identification system are as follows:

(1) Misrecognition rate: refers to the probability that two different fingerprints are incorrectly recognized as the same fingerprint;

(2) Rejection rate: refers to the fact that two different fingerprint samples of the same finger cannot match, that is, the probability of being considered to be from different fingers;

(3) Equal error rate: the value when the first and second errors are equal;

(4) Registration time: the time required from the fingerprint collection to the completion of the fingerprint feature submission;

(5) Matching time: the time required for two fingerprint samples to perform a comparison match;

(6) Size of the template feature: the storage capacity of the fingerprint feature extracted from a fingerprint image;

(7) Size of allocated memory: The amount of memory that the computer system needs to occupy at various stages of fingerprint recognition.

After the image is captured into the system by the fingerprint acquisition device, we need to evaluate the quality of the captured fingerprint image. If the quality of the image is not up to standard, it will have an impact on the later stage. Therefore, the fingerprint image needs to be evaluated. Currently, there are several methods for evaluating the quality of fingerprint images:

(1) Calculate the signal to noise ratio of the image:

This method refers to finding the ratio of the signal to the variance of the noise. First, the local variance of all the pixels of the image is calculated. The maximum value of the local variance is set as the signal variance, and the minimum value is set as the noise variance. The ratio of these is determined, and then converted into a dB number. Finally, the empirical formula is used to correct. This method performs generally in terms of efficiency.

(2) Count the number of detail points of the fingerprint image:

Identify and count the number of minutiae points in the fingerprint image. It is judged by the number of the number whether the quality of the fingerprint image is within the acceptable range. This method is theoretically feasible, but it is not efficient because it first needs to preprocess the fingerprint and extract the minutiae.

(3) Visual visitor observation:

The method is based on the visual evaluation process and the guest observation degree, and uses the set evaluation parameters to evaluate a comprehensive result of the quality of the fingerprint image. This method can make a good judgment on the quality of the fingerprint image from the whole. However, from a local point of view, the texture analysis of the fingerprint lacks the judgment of the fingerprint direction information.

(4) Calculate the fingerprint image direction information:

Starting from the local features of the fingerprint image, the quality of the fingerprint image is determined by combining the global features of the fingerprint. Whether the image is acceptable is determined by detecting the effective area and sharpness of the image. The specific method is: first, determining the foreground block and the background block by calculating image direction information; then, by comparing the ratio of the foreground block and the background block, determining whether it is a partial finger; again, judging by the size of the contrast of the image block is Fingers or wet fingers (dry fingers have a higher contrast and wet fingers have a lower contrast).

After the fingerprint image quality is qualified, the image needs to be gradated, that is, the fingerprint image is equalized, the image gray level is equalized, and the image is normalized. After these are completed, the image needs to be segmented according to certain algorithms and requirements. That is to say, the quality of the fingerprint image is very poor, and the image area that cannot be processed in the later stage is distinguished from the effective area, so that the post-processing is concentrated on the effective area, and the feature extraction precision is provided, and the processing time is reduced. At present, the commonly used segmentation methods are as follows:

(1) Segmentation method based on pattern:

The fingerprint area and the background area are distinguished according to the direction of the texture on the image, and then divided according to different areas. If the texture lines of the fingerprint are not continuous, the grayscale of the image is difficult to estimate correctly, or some areas change sharply, this method cannot be effectively segmented.

(2) Segmentation methods based on image local gray mean, local standard deviation and local consistency:

The gray level mean, standard deviation and consistency of the local area of the fingerprint image are used as features, and then the linear classification is used to segment the fingerprint image. The consistency of the partial images shows the texture trend of the partial images, but these features cannot be effectively represented for the blurred regions.

(3) Multi-level segmentation method:

That is, the fingerprint image is divided into multiple levels, and the range of the segmentation is reduced step by step. For example, the first level divides the background area of the image, the second level divides the blurred area in the foreground area, and the third stage segments the unrecoverable area from the blurred area.

(4) Dynamic threshold segmentation method:

The threshold is automatically adjusted according to the local grayscale contrast of each sub-block, and the segmentation is performed based on the variance of the pixels. The method is simple, fast, and has a good segmentation effect. Specifically: dividing the image into sub-blocks that do not overlap; calculating the average gray level and gray-scale variance of each sub-block; calculating the difference between the maximum value and the minimum value of the variance; defining a dynamic threshold value, and dividing the image; smoothing operation , remove the isolated block.

Fingerprint image enhancement is to change the blurred fingerprint texture more clearly. For example, the broken fingerprint lines are connected to separate the connected lines, and in the process, the original fingerprint image structure needs to be maintained to make the image It is easier to extract feature information. Currently, there are several fingerprint image enhancement methods:

(1) An image enhancement algorithm using a smoothing operator from the ridge direction and an enhancement operator in a direction perpendicular to the ridge line. This algorithm is theoretically quite correct, but it is difficult to estimate the ridge width and the filtering parameters. If the parameters are estimated incorrectly, the ridges will be contaminated and the fingerprints with creases on the ridges will be biased.

(2) Fingerprint image enhancement algorithm based on Gabor filter. This algorithm filters before using the previous method. Dividing the fingerprint image into different regions effectively weakens the noise perpendicular to the direction of the dominant ridge and improves the reliability of directional information extraction.

(3) Fourier enhanced post filtering method. Based on the consideration of time and processing effects, the Fourier transform is first used to enhance the fingerprint image, and then the filter is used to repair the lines of the fingerprint image. Specifically: first, multi-level segmentation of the recoverable region block, the block pixel is changed to a complex form; using discrete Fourier transform to filter out frequency band noise of too high or too low frequency; using a directional filter to eliminate fingerprint break and fork even.

Extracting on the basis of refining the image

First, the fingerprint image needs to be refined, the fingerprint line is thinned, and then the connection point in each of the eight pixels on the line is analyzed to determine the type and position of the pixel, and the analysis is performed. The line segments connected by the pixels determine the direction of the points, and then extract the feature points. The advantage of this method is that the principle is relatively simple and easy to implement; the disadvantage is that a large number of pixel points need to be refined, and the time is slow. When the image quality is not high, the refinement process will generate many impurity items.

Extract directly from the original grayscale image

Using the fingerprint pattern, the fingerprint line is tracked on the grayscale image. Each time a certain length is tracked, the position of the line is determined according to the projection extreme value of the image. When the endpoint and the bifurcation point are encountered, the projection process is automatically performed. termination. The advantage of this method is that it has higher efficiency and precision; the disadvantage is that it is complicated to implement, requires a lot of calculations, and when the image quality is not high, the obtained pattern may be unreliable, resulting in deviation of the traced lines. .

Fingerprint image matching refers to comparing the fingerprint features extracted by the current fingerprint image with the features in the pre-pre-existing fingerprint database to determine whether the two fingerprint features are consistent, that is, whether they are from the same finger. In order to avoid interference of some factors, such as deformation, false feature points, feature point position error, etc., it is necessary to design an accurate and effective matching algorithm. Currently, there are several ways:

(1) Based on the point pattern matching algorithm.

Most current algorithms are based on the characteristics of the minutiae to match. The matching is divided into the following types: the matching based objects can be divided into 1 to 1 matching and 1 to many matching; the matching based adaptation degree can be divided into elastic matching and rigid matching.

(2) Based on texture pattern matching algorithm.

Firstly, the effective area segmented by the fingerprint image is meshed, and then the Gbaor filter is used to process the ridge region from 8 different directions of the pixel, and the global information and local information of the fingerprint are obtained and converted into a feature information, and finally Compare the difference between the current fingerprint image and the corresponding feature information of the image in the database. The algorithm can solve the difficulty of image matching with poor quality and difficult to extract regional details. However, this method requires a large number of operations for each pixel, and it cannot handle the matching of fingerprint images with relatively large deformation.

With the rapid development of biometrics, users often have a question: What are the most appropriate choices for biometrics?

In addition to the fingerprint recognition that has been discussed above, there are also common face recognition, iris recognition, which technology is better?

In general, there are three ways to authenticate personally:

1) What you have, such as NFC phones, smart cards; 2) What you know, such as PIN, password; 3) Finally, you.

Cards, tokens, PIN verification and other technologies can only guarantee that the information contacted is correct, but it is difficult to guarantee that this person is real. Biometrics brings a verification loop between people and behavior, and another benefit is convenience. You won't lose it, forget it or share it with others.

The end user experience is as important as the quality of biometrics, with three influencing factors:

1) Undoubtedly, hardware product quality is a key factor in image input quality. Choosing a reliable hardware product is the basis of biometric identification; 2) Biometric identification algorithm determines the result of biometric verification, and is also an important influence of speed and performance. Factors, especially in the era of big data, are particularly important; 3) practicality is an influential factor in the selection process that is often underestimated.

Various biometric technologies have their advantages and limitations.

Fingerprint is one of the most widely used and mature technologies and the starting point for biometric applications. The fingerprint identification price is relatively cheap and functional, and it is also a very reliable verification method. However, if you need more advanced security options, fingerprint + smart card or fingerprint + password is a good choice. The low-level security options are generally applied to mobile phones or tablets, and interact with users through sensors. However, fingerprint recognition still has a large limitation on some difficult fingers (peeling, etc.) and recognition conditions (wet fingers).

Iris recognition has always been hailed as the most accurate and safe way to identify creatures. Therefore, the cost of iris recognition is relatively high, and requires professional hardware products. However, with the maturity of iris recognition technology, the current cost of iris recognition is no longer a "disappointing" state. Manufacturers have even developed sophisticated and compact iris modules for mobile terminals, which are cost-effective.

Venous identification is followed by a more secure and accurate biometric authentication method that provides a high-quality authentication method similar to fingerprint recognition, and can also interact directly with the user.

Face recognition is non-contact recognition, very friendly and convenient, and is a better way to use the camera. It can be used in a wide variety of environments, including construction sites, mobile devices, website logins, and even without special hardware support.

Speech recognition is also a very convenient way to identify. For mobile devices, this is a quick and easy verification solution. However, the quality control of voice is one of the key factors affecting its identification and development. Its security is also worrying, but it can still make a big difference in call center and customer service.

New technologies such as electronic signatures, gait recognition (walking contours), and ear pattern recognition are still in a wait-and-see state.

There is no uniform answer to which biometric method to choose. It depends on what the user is doing, but there are general principles.

1. Low-cost solutions still tend to fingerprint, sometimes with a smart card or password.

2. Mobile solutions usually use face and voice, fingerprints are also very popular, but now iris recognition has gradually become an important means of competition for mobile phone manufacturers.

3. If you want non-contact recognition, it tends to face recognition or iris recognition.

4. Secret surveillance tends to face recognition, speech recognition and gait recognition.

5. Large-scale projects often use fingerprints or irises. For example, Chinese ID cards will be included in fingerprints, while Indian biometric items include iris recognition.

6. Projects with higher security levels tend to be iris or finger vein recognition, and often combined with cards or passwords to form multi-factor verification.

7. If the certification conditions or the environment is bad, then the iris is preferred.

Chat Online 编辑模式下无法使用
Chat Online inputting...
Too many consultations now, we will respond via call or email in 1 weekday.