TW476917B - Hand features verification system of creature - Google Patents

Abstract

A hand geometry-based verification system of creature that utilizes the overall feature verification system process of creature can be categorized as two stages, namely enrollment and verification. First of all, a specially developed palm image capturing device is used to get high-quality image data and then the geometry features of palm are used to indicate the locations of fingers with respect to palm center area to obtain the geometry features of hand. In addition, the palm print data of palm center area and on fingers can be extracted. During the enrollment stage, after the palm geometry feature data is analyzed, the center point and variance distribution range can be calculated for the calculation of hand geometry deviation. The palm print data goes through the main axis analysis and the quantification of generalized learning vector to obtain the palm print reference vector based on individually different resolution. Bootstrap technique is used to generate additional palm print data to acquire the best positive Boolean function, thereby automatically integrating the correlation of each resolution of palm print data to obtain the optimized verification result. During verification stage, image input goes through the feature extraction procedures and so forth. If the deviation of hand geometry is too big, it can be considered as an intruder, or compared in the palm print verification process to verify the identity. The experimental data proves that the present method is effective in terms of result.

Description

476917 A7 B7 V. Description of the invention (PA88Q369.TWP-3/25 [Technical Field 10 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs of the People's Republic of China 20 printing knife breakage confirmation system, especially the age-specific use of each person's body to feed the material㈣ For those who collect or act, their towels and palm-shaped or palm-printed features are most suitable for making on various systems on the Internet. Therefore, only a small number of beauties in the office can represent their personal identity and make the data on the network transfer quickly. It can be faster. The present invention focuses on the palm-shaped and palm-printed biological characteristics, and confirms the identity of the individual on the Internet. The Internet users themselves' rights and interests. "[Previous technology] In recent years, biological characteristics have gradually become everyone The objective is the focus of the purpose. Using the individual's unique biological characteristics as personal identification confirmation, and more protection of everyone's rights and interests, they have gradually replaced the status of password protection (⑽㈣Identificati〇 Ν-ΡΙΝ). For a long time, although Mima A method that has been used to protect personal data, but due to forgotten passwords or stolen passwords, personal data flows out, bank accounts Deposits are subject to heavy losses, heavy personal data, and heavy social costs such as setting. The solution is to avoid using passwords that are too simple and early, such as birthdays and ID numbers, and you need to change the passwords frequently. 'Often there will be a dude who even forgets his own password, and the biological feature overcomes this shortcoming. It eliminates the trouble of memorizing the password and carries it with him. In addition, it has the advantages of uniqueness and difficult to counterfeit.' 疋A very good way to protect the security of personal rights. The biometrics confirmation system is to use the unique characteristics or behaviors of each person's body, such as voice prints, palm prints, signatures, eye iridescence, fingerprints, palm shapes, Facebook and other information. , For personal identification. In recent years, many --- 丨 --------. «Clothing (please read the precautions on the back before filling out this page) Order ----line--nnnn This paper size_ (210 x297 mm

A7 B7 V. Description of Invention (^) ---, ^ PA880369.TWP -4/25 Researchers have also devoted a lot of effort to a wider range of security systems, and the guide lists one for each feature table). Committed to seeking better algorithms and pushing to protect personal rights. Many articles report their advantages and disadvantages, we summarize them (table confirmation technology

g Zhiyi is easy to obtain and imitate with loyalty fingerprints, so it is not suitable for popularization | 境 # 果 严 掌 形 面 表 5 Confirmation rate, convenient for users, has been closed and the Olympic gate ^ Users are most convenient, no contact Operation f 'is suitable for access control and bank ATMs. Low uniqueness, low recognition rate, difficult to extract facial features Signature dynamic features are not easy to imitate, convenient for users, suitable for office automation systems, PDA, etc. Significant variability in voice Signature Convenient for users, non-contact operation, and the only confirmation application suitable for use over a telephone line. It is more susceptible to environmental interference. Restriction pattern, confirmation rate, unique, [± high, can be used for security control of sound secrets. Serious interference to users Basically, various intelligent personal identity verification technologies have their advantages and disadvantages. ^ ... ,,, and good or bad, all depend on the purpose and environment of the system. Similarly, Ken Phillips also targets biological Features analysis and comparison of application systems on the Internet. The British AFB Society (Association f0r Biometrics) has proposed five major steps for the degree of safety requirements as a selection of biometric confirmation / 〇y 丄 7 A7 B7 PA880369.TWP-5/25 V. ίο 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 The basis of the invention (ingenious) system. It can be seen that if the construction cost is cheap and can be uploaded on the network for confirmation work theory, the palm shape and palm print confirmation system is the most appropriate. The present invention focuses on the palm shape and palm print confirmation system. Many research scholars have also proposed many palm-shaped feature comparison methods to improve the confirmation rate of personal identity confirmation. For example, Zhang and Shii used the datum point invariance principle and line segment feature comparison skills to perform palm print confirmation. First, they will Printed on the ink card, and then scan the ink card with a scanner, and take the ink image with a size of 400x400. These ink images can be provided to police units as evidence for criminal evidence. Scanning is therefore less suitable for general online identification systems. Kung β α /. Once designed a “decision-based neural networks” (DBNN) recognizer for face recognition, and extended this recognizer to palmprint confirmation. Joshieia /. Using a CCD camera to obtain the contour of a finger (middle finger), a set of 472-bit "wide line integrated profile (WLIP)" data was generated. And use the "normalized correlation function" (normalized correlation function) to calculate the difference between the input sample and the reference sample. * It can be seen that the above-mentioned conventional methods still have many shortcomings. They are not a good designer and need to be improved. In view of the various shortcomings derived from the above-mentioned conventional feature recognition system, ‘is anxious to improve and innovate’, and after years of painstaking research, finally successfully developed this palm biometric recognition system. [Objective of the Invention] The purpose of the present invention is to provide a palm biometric identification method. The size of the paper is covered by Chinese National Standard (CNS) A4 (210 X 297 mm). --- P ------ 1- ---------------- ^ (Please read the notes on the back before filling this page) 476917 A7 B7_PA88Q369.TWP-6/25 V. Description of the Invention ([[/) System, The present invention focuses on integrating palm-shaped and palm-printed biological characteristics to confirm an individual's identity on the Internet and protect the rights and interests of Internet users. Because the use of palm-shaped or palm-printed characteristics is most suitable for various systems on the Internet, Only a small amount of data is needed to represent personal identity, so that data on the Internet can be transmitted faster than 5 speeds. A second object of the present invention is to provide a palm biometrics identification system, which uses palm imagers to obtain high-quality image data, and then uses the geometric characteristics of the palm to mark the positions of the fingers and the palm area. The geometric features of the palm shape, in addition to extracting palmprint data with multiple resolutions in the palm area and on 10 fingers. Another object of the present invention is to provide a palm biometric identification * system. The entire system can be divided into two phases: the registration phase and the confirmation phase, and during the registration phase: the palm-shaped feature data can be obtained after analysis The center point and its variation distribution range are used to calculate the palm feature error value of 15. The palm print data is quantified by the main axis analysis and the general learning vector to obtain the palm print reference vector of each person at different resolutions. Using bootstrap techniques, additional palm prints are generated Data, find the best positive Bollinger function, and automatically integrate the relationship between each resolution palmprint material to get the best confirmation result. In the confirmation stage: the input image is subjected to feature extraction and other steps. If the palm shape error is greater than 20, it can be regarded as an intruder. On the other hand, the identity of the person can be determined only after the palm print confirmation process is performed. [Technical content] Hao has the above advantages of the palm biometrics identification system, which uses two physiological characteristics of palm shape and palm print for personal identity verification. The paper size is applicable to China National Standard (CNS) A4 specification (210 X 297 mm) (Please read the notes on the back before filling out this page) Order · ·-丨 · Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs 476917 Printed by the Employee Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 __ _B7_PA880369.TWP -7/25 V. Invention description (7) includes two phases of registration and confirmation. During the registration phase, the palm data of each person will be collected, feature extraction and analysis will be performed to generate a personal comparison template and related parameters; and In the confirmation phase, the newly input palm samples are compared and confirmed. The entire system architecture is divided into six modules: 5 including image capture, wavelet-based image segmentation, hand geometry-based enrollment, and palmprint feature registration (palm) -print enrollment), hand geometry-based verification, and palm-print-based verification. First of all, we will ask the user to put the right palm on a platform, and input the palm image to the computer through the CCD camera. And use the wave element-based image segmentation technology, plus the geometry between the palm and the knuckles. Relationship, automatically indicate the position of the fingertip of the finger and the palm area, and then extract 11 palm features and several palm print features with different resolutions through the palm shape / palm print feature extraction step. 15 Palm-shaped registration module for analysis and training, and then preliminary confirmation by the palm-shaped confirmation module, while the f-print feature is transferred to the palm-print registration module and group to obtain a comparison template to facilitate comparison with the palm-print confirmation module. For exact confirmation, the palm / palm confirmation principle is described as follows: The palm feature vector is' accounted for-palm feature error value in the palm disc recognition module. If the value is greater than a certain thresh value, then It means that the input sample is not owned by the person. Otherwise, it may be owned by the person. It is still necessary to confirm with the palm print feature and perform the final confirmation 'to complete the confirmation. [Simplified diagram _] (Please read the notes on the back before filling this page).% Line. Please refer to the following detailed description of a preferred embodiment of the present invention and its

476917 Α7 Β7 PA880369.TWP-8/25 V. Description of the invention (5 10 The drawings will further understand the technical content of the present invention and its purpose and effect; The drawings related to this embodiment are: / Figure 1 is the palm of the present invention Figure of the architecture of the biometric identification system; Figure 2 (a) is the third view of the palm image viewer; Figure 2 (b) is the external view of the palm imager; ... Figure 3 is the palm image view; 'Figure 4 ( a) is the basic element signal cutting diagram (original signal); Figure 4 (b) is the basic element signal cutting diagram (when Scaie = 2); Figure 4 (c) is the basic element cutting signal diagram (Dangshi = 4) Figure 5 is a palm-shaped feature view; Figure 6 is a palm-print feature view; and Table 1 is a personal intelligent personal identification technology that is currently in development and has been developed. · [Representative Symbols of Main Parts] 11 Image Mining Module 13 Leather-shaped feature registration module 12-wave basic image segmentation module Η Palmprint feature registration module (please read the precautions on the back first and then this page) 1 £ -H ϋ I ί · _ϋ n 口 VI m · m nn line · Economic Printed by the Ministry of Intellectual Property Bureau's Consumer Cooperative 15 15 Confirmation. [Preferred Embodiment] Please refer to ®-, the architecture diagram of the palm bio-special wealth recognition system of the present invention. 'It can be seen from the figure that the present invention uses the two physiological characteristics of the palm shape and the palm print' for personal identification. , Which includes two stages of registration and confirmation, the bribe and towel will collect the palm of each person's palm, and enter: mold 16 palm print feature confirmation module

476917 Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs A7 ___ _B7_PA880369.TWP-9/25 V. Description of the invention () Extraction and analysis, and then generate a personal comparison template and related parameters; and the confirmation phase is for new input Palm samples for comparison and confirmation. The entire system architecture is divided into six modules: including image capture module 11, wavelet-based image 5 segmentation module 12, and hand geometry-based enrollment. Module 13, a palm-print enrollment module 14, a hand geometry-based verification module 15 and a palm-print-based verification module 16. First, we will ask the user to place the right palm on a platform, input the palm image to the computer through the CCD camera, and use wave element-based image segmentation technology, plus the geometric relationship between the palm of the hand and the knuckles. Automatically mark the position of the fingertip of the finger and the palm area, and then extract 11 palm-shaped features and several palm-print features with different resolutions through the palm-shaped / palm-print feature extraction step, in which palm-shaped feature data is submitted to palm-shaped registration Module, perform analysis and training I5 practice, and then make preliminary confirmation by the palm shape confirmation module, and the palm print feature is transferred to the palm print registration module to obtain a comparison template to facilitate more accurate confirmation with the palm print confirmation module. The palm / palm confirmation principle is described as follows: The palm feature vector is calculated as a palm feature error value in the palm confirmation module. If the value is greater than a certain threshold (threshold), it means that the input sample 20 is not I own it, otherwise, I may own it. I still need to confirm it by palming the 4-inch levy and perform the final disc recognition to complete the confirmation. The following introduces the six modules in this system: ^^ Palm image manipulation device design (design of hand shape / palm-print __ -9- This paper size applies to the Chinese National Standard (CNS) A4 specification (210 χ 297 Public love)-^-ϋ βϋ n ϋ ϋ ϋ II nn · n Βϋ lnn nfl an. Ί Ί Ί n ϋ nn (Please read the notes on the back before filling this page) Line. 476917 A7 V. Description of the invention ( ^ 5 10 15 20 image capture ～ like α ”'is related to the quality and quality of the entire system, which will not only get better confirmation results, but also save a lot of effort = == meter, the images used in the present invention are all used by us The self-developed hand sheep camera was obtained, as shown in Figures 2 and 3. Figures == ::: Figure 2 _ are placed one by one for their appearance: to find the finger in the image in Figure 3. She placed eight, soft 4 '. Asia, and voluntarily moved the palm area households! Households 2 households 4 弋 showed that because everyone has a different palm, and each person's palm shape is different, Because we need a stable method to find the palm area, the strategy we use is the "wave element basis" ( wavdet_based) segmentation method, it can: indeed :: ten different edge point (edgep〇int), and indicate the position of the fingertips, consider the image of Figure 3, because the position of the six small pillars is fixed, can be manually pre- Enter the position coordinates, and use the wave element conversion for the gray-scale value contours of square private type (less = a + 70), and calculate the low-frequency and high-frequency signals with multiple resolutions, as shown in Figure 4. It is a very good method for image segmentation. The edge points are all located on the same frequency signal (in the present invention) = 4). Therefore, it is easy to find each branch. Finger cutting line fairy, CD, sand, the same, the other three cutting lines can also be the same -10

Zhang Zhihan's Chinese specifications⑵0 x 476917 A7 B7 V. Description of the invention (PA880369.TWP-11/25, percussion long stalk ° Here, 'we assume that the middle line of the finger is in the middle of the two secant lines: purpose' = Using geometric formulas, calculate the three fingers: the index finger and the centerline of a ^ 9: Α, Zn2, ζ3 'According to our experience, no matter how the palm is placed, in the palm area W4P3, the line segment is miscellaneous 2ρ4 5 :: and In the same direction, calculate "2 and ⑽, 3", 2) will intersect r, and household A is a straight line that passes through and is perpendicular to Yibi (heart line). The points 弋 and dim sum are the intersection points of tangent lines and, respectively. Among them, 1 household 3 'In addition, the palm area household / ^ is defined as the area of the square that is 12 times larger than the line segment. Similarly, you can find 10 three-finger endpoints on the line. f --- r -------- (Please read the notes on the back before filling out this page) 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 (Hand Ge ^^ m, print Feature Extraction, Yes Given the positions of these key points, we can smoothly extract 丨 丨 palm-shaped feature vectors' and several multi-resolution palm-print feature vectors. As shown in Figure 5 = each palm-shaped feature is marked: Ku, of which the line segment is inverse And the line segment @ and the line segment @ are mutually perpendicular to each other, and are located at the positions of y? And 5/7, and the length of the line segment Qiao, both and obtained, respectively represent palm-shaped features Bu 2, 3, the same as the 'feature 4 ~ 9, The palm-shaped features of the index finger and ring finger, respectively. In addition, the lengths of the line segments W (feature 10) and @ (特) represent the palm-shaped features of the area of the palm area, where both are 0.25; ^. ， &Quot; / 、 Next, We can follow the procedure below, from the palm area and hand -11--order ----- thread paper size applicable to China National Standard (CNS) A4 specifications (210 χ 297 mm) points / 6917 B7 _ PA880369.TWP -1? /% Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs (Refers to the area to obtain palmprint characteristics. According to the introduction in the previous chapter, we can stably "show the palm area. Everyone's texture distribution in this area is not the same." Because the amount of data in this area is quite large, we only take out the area. The grayscale value contour of a specific line segment in the middle (as shown in Figure 6), where the points 5 仏, β, and β are located at the position of the line segment @ 中 1 / 4,2 / 4,3 / 4, and the remaining points β, , ..., can be obtained in a similar way, in addition to the gray scale contour of the angle line @ @, there is one point worth noting. The diagonal line is not considered, the reason is the way the palm is arranged, which makes The user's 1 point is outside the palm of his hand. The line segment is blended with the background. Therefore, the gray level value contour of the line segment is not considered. Consider the gray level value of one of the line segments and obtain several multiples through wavelet transformation. The resolution signal OT is taken from the wind ..., wind, ㈣, ........., _, Caab used to analyze the signal after the wavelet conversion of the two-dimensional Facebook image, and analyzed the palmprint signal from the wind, and only took it out. Fractional SF In addition, since the palm area of each person is different in size, the normalized silk degree of each resolution is corrected, so that the length of the dry heart region in each palm mosquito image can be obtained as the feature vector. In addition, three隹 The grayscale value of the finger also contains a wealth of resolution information. You can also use the 1 ^ Beicheng of the field 'can also be a 4-inch resolution.' Similarly, the line segment is reversed, and the obtained grayscale value is the same. Round 2 has been simplified and normalized into the feature vector of z. In the meanwhile, we'll obtain the feature vector of dimension K). For us, there is still too much to be desired. Still ... I think it ’s too eve (it ’s passed on the Internet, the smaller the data passed, please read the back first 10 15 20 -12-

Note matter

476917 PA880369.TWP-13/25 A7 B7 5. The invention description (u) is better 'so after a vector with a length of 10L is converted into a vector with length A: (ie ·, 10) after the main axis analysis, it can be accelerated Delivery on the web. Therefore, each sample X can be represented by ruler dimension feature vectors RFQ, RF \, ..., RFK ^ at each resolution. 5 (4) Hand Geometry Feature Registration (Hand Geometry Feature Enrollment) The use of palm feature for identity verification is a very good method to confirm the effect, and there are already commercial products, but the palm feature is often criticized as: The palm shape will change with time. Even if the amount of change is small, the confirmation rate will be reduced due to the high repetition rate of 10. Therefore, we expect to use the palm feature to make a preliminary confirmation to minimize the "false rejection rate" (FRR ) To make the system more tolerant of changes in palm-shaped features due to time, and then use palm-print feature data for accurate confirmation. The 11 palm-shaped features can be obtained from the wave element-based segmentation method and some simple geometric formulas. Assuming that a particular person has 15 training samples, the average value of the sample (/ 〇 and the standard deviation (σ ), According to these two data, in the palm verification module, the similarity between the input sample and the training sample can be calculated. (5) Hand Geometry-based Verification 20 The similarity between the input sample x and the template is evaluated by the following equation:

σ V σΐ J -13-This paper size applies to Chinese National Standard (CNS) A4 (210 X 297 mm): --- Γ ---------------- Order- ------- Line j (Please read the precautions on the back before filling this page) Printed clothing for the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 476917 A7--~ -_____ 2ί ____ PA880369.TWP-14/25 V. Description of the invention ((Please read the precautions on the back before filling this page) If the similarity d is greater than a certain threshold, input the sample "as the palm data of the intruder, otherwise, it must be further confirmed by palm print characteristics 'You can determine whether you are who you are. 5 Palm-print Feature Enrollment The purpose of this module is to find personal comparison information, which is mainly divided into two steps' 1 ^', which are general learning from Cao Generalized learning vector quantization verification (GLVQ) and search of optimal positive Boolean function (OPBF search). 10 Learning vector quantization (LVQ) is a supervised classification first proposed by Kohonen> The Bayer method is used to produce the best The test sample is also a fairly simple and fast learning algorithm. Many research scholars have proposed using learning vector quantization (GLVQ) for its shortcomings. Sato and Yamada improved to become general learning vector quantization (GLVq) in 1995. ), Using the method of 15 Sato and Yamada in the creation, to obtain a better reference feature vector of the palm print sample 'Since it is extended from the LVQ algorithm, we will divide lvQ and GLVQ 丨 J as follows: Ministry of Economic Affairs Printed by the Intellectual Property Bureau employee consumer cooperative in [3], assuming that X is a certain training sample, the LVQ2 algorithm designs the reference vector learning rule as: (ί + 1) = w. (〇-a (t ) (x ^ w. (ή) wk (t + 1) = wk (t) + a (t) (x-Wk (t)) where · is expressed as' • order in time, j, The value 0 < Λ (ί) < 1 'and the meaning is a decreasing time function. Sato and Yamade improved -14-This paper size applies the Chinese National Standard (CNS) A4 specification (210 X 297 male; f) 476917 A7 B7

PA880369TWPjJ ^ iiL 5. Description of the Invention (^) The disadvantages of the algorithm LVQ2.1 cannot converge. They redefine% as the nearest reference vector and belong to the same group as X (ie: class (χ) = class (> ^)), W2 is the nearest reference vector, but belongs to a different ethnic group (ie ·, class (x) 矣 class (i〇), and its relative distance a W is defined as 5 μ {χ) = (-d2) / (+) Where 4 = | x —wz_ | and X-wj are defined as the distance between x and π, respectively. They design a general learning vector quantization (GLVQ) algorithm, and its learning rules are designed as follows: (Please read the notes on the back before filling out this page) 10 ^ x {t + \) = wx {t) + a df d2 3μ ( dx -f d2y xo),

«— III where df_ θμ / (/ Μ) {1-/ (#, ί)} is an update of the above formula, which is printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs and printed 15 children '/ (/ ^) 二 l / ( l + e 卞) is a sigmoid function, and the value α is set to 0.001 in this paper. A detailed description can be found in the literature ρ]. In the invention, our main job is to confirm whether the input image is owned by us? Therefore, this problem can be regarded as a problem of tw〇_dass classification. There are only two ethnic center points, which are YES (Wl) and ΝΟ (νι; 2). It is independent of each other. Assume that someone X has M training samples, and randomly samples ^ samples (miscellaneous people) '. The corresponding ethnic center% and% can be taken from the training sample. 15 Line: This paper scale applies Chinese National Standard (CNS) A4 Specifications (21〇χ 297mm7 " ^ / 0917 Five A7 B7 PA880369.TWP-16/25 Printed by the Consumers ’Cooperative of the Intellectual Property Bureau of the Ministry of Economics N Description Φ The average value is obtained. In this case, my training The number of samples is relatively small, but many non-self training samples can be collected, because as long as the palm prints of non-myself can be regarded as non-my own training samples, in order to balance the difference in the number of training samples, the M training samples of X are copied # Copies, and then mixed with other non-five personal samples to become 2MM samples. As a round-robin sample of the GLVQ algorithm, after repeated calculations, it will eventually converge to a stable state. Two sets of reference vectors, one for yourself and one for others. If you input palm print on image λ, after pre-processing, palm area labeling, feature extraction, and dimension reduction, etc., If the X vector is closer to wi (ie, yi (x) < 0), the input pattern can be regarded as the identity of the person, otherwise // (X) > 〇, the input identity is regarded as an intruder. After the GLVQ process We can judge whether the identity of the input palmprint image is owned by ourselves. As mentioned in the previous section, palm wave features can be expressed into several multiples by wave element conversion and principal axis analysis. Re: solution. Resolution vector along "., Along ^ , ......;? ^, And then through the GLVQ process, the reference vector of each resolution can be obtained, and the input vector will be compared with the reference vector to calculate the Α (χ) value to determine whether the input palmprint data is himself All, but each person's confirmation effect will be different under different resolutions. Code X will have a better confirmation result at resolution household 2 and code Υ will have a better confirmation at resolution household ^. As a result, we want to be able to set One method can automatically integrate the confirmation results of each resolution according to each person's palmprint characteristics, and obtain a better confirmation effect. Therefore, we use a positive Bollinger function to integrate the judgment results of each solution. First of all, we need additional characteristic data to evaluate the effect. -16-10 15 20 This paper size is applicable to the Chinese National Standard (CNS) A4 specification (21〇χ 297 mm) (Please read the notes on the back first (This page) Binding --------- line ί _ A7 _ B7 one " " "-", 丨 " · '" ----- PA88Q369.TWP-17/25 V. Description of the invention (A better confirmation result, so we will randomly rotate (-5 degrees ~ +5 degrees), zoom in or out (_3% ~ + 3%), χ Directional displacement or less direction (-2pixel ~ + 2pixel), and then extract the multi-analyzed feature vectors according to the palm shape / palm print feature extraction step, and process them to obtain the extra 5 evaluation data. Here we assume the evaluation data and Confirmation results for future test data will be similar. The positive boolean function (PBF) has been successfully applied to stacked filters. It is mainly used for noise removal and edge detection. Each positive boolean function represents a filter, and the best The stacking rate wave filter 10 is a positive Bollinger function with the smallest absolute error value among the many Bollinger functions. It is applied to noise removal. The noise image is filtered by the best rate filter to remove the noise. The error Z between the generated image and the predetermined image is small. The present invention uses this concept to automatically integrate the relationship of each resolution to obtain the best confirmation rate. Prior to this, we identified 2 of the 15 questions that turned into optimal positive Bollinger functions. Considering each resolution, it can be regarded as a random variable, and each training sample can be regarded as a realization. Under the supervised learning method, each training sample is known; whether it belongs to my palmprint data , Is regarded as a predetermined result, and after each sample of M resolution data passes the positive Bollinger function, a 20 confirmation value will be generated. Our goal is to generate all the training samples by positively distributing I and then generating The minimum absolute error between the result of and is the smallest. This positive Bollinger function is the integration function we need. Through this Bollinger function, we can get better confirmation results. , (Please read the precautions on the back before filling this page) ----- Order --------- Line Employees of the Intellectual Property Bureau of the Ministry of Economic Affairs print f-J cooperatives -17-

476917 A7 —One ----------- PA880369, TWp, 18/25 V. Description of the invention The operation flow of the present invention includes two stages: Registration stage confirmation stage • Enter the user identification name and step one: Enter the user identification name and collect M personal training samples and mn non-self training samples. Step 2: Use the self and non-self training samples to perform wave elementary image segmentation. Step 3: Extract palm feature data and several multi-resolution palm prints. Feature data Step 4: Generate palm template (", (7) ----- ^ ·: · Step 5: Generate several multi-resolution palm print templates Step 6: Universal learning vector quantization fine-tune palm print template (0〇 , 1, Aj, 〇02, called '2 ......) Obtaining image Step 2: Input image as wave elementary image segmentation Step 3: Palm feature data extraction Step 4: Calculate palm feature similarity If it is greater than a certain threshold value, it is regarded as an intruder; otherwise, perform step 5 and step 5: several multi-resolution palm print characteristics data. Printed by the Consumer Cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs. Step 7: Search for the best positive Bollinger function

〇 PBF Step 6: Compare multi-resolution palm print templates Step 7: Integrate several multi-resolution confirmation results, if it is greater than a certain threshold value, then it is regarded as an intruder I _-18 _ National standard of this paper country ^ (cns) Regulation a4 ^ 476917 A7 B7 PA880369.TWP-19/25 Employee Consumption Cooperatives Printed by the Intellectual Property Bureau of the Ministry of Economic Affairs V. Experimental Results Here we have designed some experiments to prove what we have mentioned The method is sufficient to provide a good confirmation effect. First, we collected 10 30-5 palm-shaped / palmprint image data of each person. In this experiment, we will select 10 of them as training samples, and the other 20 The image data is used as a test, and the false rejection rate (FRR) test is performed on 20 images of a certain identifier X, while the other nine people have 20 images each and a total of 180 images are subjected to a false acceptance rate (FAR) ) Test, the final average confirmation rate is 10: FAR = 1.45% and FRR = 3.1%. [Features and effects] · The palm biometric confirmation system provided by the present invention has the following advantages when compared with other conventional technologies: 1. The palm biometric confirmation system of the present invention focuses on the integration of 15 palm shapes and Two biological characteristics of palm print to confirm the identity of the individual on the Internet and protect the rights and interests of Internet users; because the use of palm print or palm print features is most suitable for various systems on the Internet, only a small amount of data is required to represent the individual Identity, which enables faster data transfer on the Internet. 2. The palm biometric identification system of the present invention uses a palm to take 20 cameras to obtain high-quality image data, and then uses the geometric characteristics of the palm to mark the positions of the fingers and the palm area, thereby obtaining the geometry of the palm shape. Features, in addition to extracting palmprint data from palms with multiple resolutions on the fingers. Third, the palm biometrics identification system of the present invention is in the whole system -19- (Please read the precautions on the back before filling this page) Order: --line 'This paper size applies to China National Standard (CNS) A4 specifications (210 X 297 mm) 476917 Α7 Β7 PA880369.TWP-20/25 V. Description of invention (10 15 The printing king of the staff consumer cooperative of the Intellectual Property Bureau of the Ministry of Economic Affairs can be divided into two phases: · Registration phase and confirmation phase, After the registration of the palm white feature data, after analyzing the palm shape feature data, the center point and its variation "knife cloth range" can be obtained to distinguish the palm feature error value. The palm print data is quantified by the main axis = analysis and general learning vector. Obtain palmprint test vectors at different resolutions for each person, and use bootstrap techniques to generate additional palmprint data, find the best positive Bollinger function, automatically integrate the relationship of palmprint data at each resolution, and find the best Confirm the result. In the confirmation phase: the input image is subjected to feature extraction and other steps. If the palm shape error is too large, it can be considered as an intruder. Determine whether you are your own identity. The detailed description above is a specific description of one of the feasible embodiments of the present invention, 'but this embodiment is not intended to limit the scope of the patent of the present invention.' Implementation or change_ should be included in the scope of patents in this case. * = The 'this case is not only innovative in terms of technical ideas, but also can be compared with the above two effects, should be fully in line with the nature and progress The statutory invention patent elements, the difference between the provisions of the law to make this fortune "«, ㈣ correct, ^ miscellaneous pages

This paper Shoulder (21〇χ 297mm) 476917 PA880369.TWP-21/25 A7 B7

Q V. Description of the Invention (θ) References 1. D. Zhang and W. Shu, nTwo novel characteristics in palmprint verification: Datum point invariance and line feature matching ", Pattern Recognition, vol. 21, pp. 691- 702, 1999. 5 2 · AFB, "A Five Step Guide to Selecting a Biometric System",

Association for Biometrics, ftp from http: // www. Afb. Org. Uk / pub / 5 staps. Html 3. A. Sato and K. Yamada, `` Generalized learning vector quantization, '' Advances in Neural Information Processing 8, 10 Proceedings of the 1995 Conference, pp. 423-429, MIT Press, Cambridge, MA? USA, 1996. 4. AI Gonzalez, M. Grana and AD Anjou, `` An analysis of the GLVQ algorithm '', IEEE Trans.〇 n Neural Networks, vol. 6, no. 4, pp. 1012-1016, 1995. 15 5. K. Etemad and R. Chellappa, `` Discriminant analysis for recognition of human face recognition / Journal of Optical Society America, vol. 14, no. 8, pp. 1724-1733, August, 1997. 6. RL Zunkel, " Hand geometry based verification " Biometrics: Personal Identification in Networks Soceity, ed. AK Jain, R. Bolle, 20 and s. Pankanti , 1999. 7. K. Phillips, " Biometric identification looms on landscape of network log-ins: High-end technology is becoming more affordable / 1 PC week, March, 1997. http: // www. zdnet.cQin / pcweek / reviews / Q324 / 24biotab.html -21-This paper size applies to China National Standard (CNS) A4 (? J0 x 297 mm) (Please read the precautions on the back before filling this page) Economy Printed by the Ministry of Intellectual Property Bureau's Consumer Cooperative

Body-· nnnn I n — 1 II n ϋ nn ϋ I n ϋ nnn 1 · II -I 476917 A7 ----- ----_ B7_PA880369.TWP-22/25 V. Description of the invention &S; Y_ Kung, SH Lin, and M. Fang, `` A neural network approach to face / palm recognition, M International Conference on Neural Networks, pp. 323-332, 1995. 9. DG Joshi, YV Rao, S. Kar? V. Kumar , and R. Kumar, 5 MComputer-vision-based approach to personal identification using finger crease pattern, '' Pattern Recognition, vol. 31, no. 1, pp. 15-22, 1998. Employee Consumer Cooperatives, Bureau of Intellectual Property, Ministry of Economic Affairs Printing 2 2 This paper size applies to China National Standard (CNS) _A4 (210 x 297 mm)

-s Please read first-Note on the back A

Claims (1)

幷 / 0917 Scope of patent application 10 15 Printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 20 A palm biometric confirmation system, including an image capture module, which can simplify the verification of the image obtained through the palm camera Work, can further improve the effectiveness and efficiency of the entire system confirmation:: wave element base · 彡 image split dragon group, wave element theory for the material signal cutting L sacrifice fruit can be divided into knuckles, this step is Palm-shaped / Fuprint feature extraction before processing;-Palm-shaped / Palm print feature extraction module 'Feature extraction is a very important pre-processing in the pattern recognition system. In the present invention, only certain specific line segments in the palm area are extracted. Gray level value, and transforming it through the number of eigenvectors of different resolutions .. Each eigenvector will be transformed by the principal axis analysis (PCA). It can not only reduce its feature: the dimension of the quantity ( dimemicmality) 'and will obtain the principal component vector of the feature (in this way, not only can maintain the characteristics of the original palmprint image') but also reduce the complexity of the feature, simplifying the following Module; A palm-shaped feature registration / confirmation module 'The feature is to confirm the identity of the individual. The most direct method is because the palm-shaped data is small, it can be quickly matched', but because of high repeatability and easy to change over time, so The use of the Shanghai confirmation method can save the confirmation time and improve the efficiency of the entire system. It is a palm print feature registration / confirmation module and group. The palm print feature is like fingerprint information-it can represent personal identity. This module uses multiple resolution patterns. To confirm the features, and use general learning vectorization and = material palm good (please read the notes on the back before this page)-I 1 II I-^ · IIIIIIII ♦ 1--23-'The paper size is applicable to China Standard (CNS) A ^ Specification (21〇X 297mm), / υνί / g ^ 380369.TWd, printed by the Consumer Cooperatives of the Intellectual Property Bureau of the Ministry of Economic Affairs 6. Application scope of patents 1 = number of integrated multiple analytical financial Grain, it is expected that with the above module, the user will place the right palm on a flat A 5. The r: D camera 'inputs the palm image to the computer and:-m basic image segmentation technology, coupled with the geometric relationship of the hand, automatically extracts the position of the finger points /, A, and the palm area to indicate dry shape / palm print feature extraction Step 'Extract a few palmprint features with different resolutions'. Among them, the palm-shaped feature Beco 乂 is twisted by a dry-shaped registration model, analyzed and trained, and then confirmed by the confirmation module for preliminary confirmation, and the palmprint feature is submitted to palmprint registration. The branch group obtains a comparison template to facilitate more accurate confirmation with the palmprint confirmation module. The principle of palmprint confirmation is described as follows: Take the levy vector in the palm confirmation module to calculate a palm-shaped feature error. The value 'even if the value is greater than a certain -threshold value, then the surface dry input sample' is not owned by myself, otherwise, it may be owned by the owner, and it still needs to be confirmed by palm print special wealth for the final confirmation. . A kind of palm biometric stone bird identification system, which uses two physiological characteristics of palm shape and palm print to perform personal identity verification, including registration and confirmation of two-stage operation. During the registration phase, each person's hand will be collected = The poor material 'performs feature extraction and analysis, and then generates a personal comparison template and related parameters; and the confirmation phase performs comparison and confirmation on the newly input palm sample. The palm biometrics identification system described in item 2 of the scope of the patent application -24- This paper size applies to China's national (CNS) A4 specification_⑵〇 χ 观 — 10 15 20 2. 3. 476917 A8 B8 C8 D8 PA880369.TWP-25/25 Sixth, the scope of patent application, the registration phase includes the following steps: Step 1: Enter the user identification name and collect M personal training samples and MN non-self training samples Step 2: Divide the wavelet base image into 5 segments of the self and non-self training samples; Step 3 · Extract palm feature data and several multi-resolution palm print feature data; Step 4: Generate a palm template; Step 5: Generate several multi-resolution palm print templates; 10 Step 6: Universal learning vector quantization to fine-tune palm print templates; Step 7: Search for the best positive Bollinger function. 4. The palm biometrics confirmation system as described in item 2 of the scope of patent application, wherein the confirmation phase includes the following steps: Step 1: Enter the user's identification name and obtain the image; 15 Step 2: Enter the image as a wave Meta-basic image segmentation, Step 3: Extract palm feature data; Step 4: Calculate palm feature feature similarity greater than a certain threshold, consider as an intruder otherwise go to Step 5; Step 5: Several multi-resolution palm print feature data 20 Step 6 · Compare multiple resolution palm print templates; Step 7: Integrate multiple multiple resolution confirmation results if they are greater than a certain 'critical value, then they are considered intruders, otherwise they are legitimate users. -25- This paper size is in accordance with Chinese National Standard (CNS) A4 (210 X 297 mm) (Please read the notes on the back first to write this page)