TW201203131A - System and method for hand image recognition - Google Patents

Abstract

A system and method for hand image recognition is disclosed. The system and method for hand image recognition includes an image capture, a feature capture, a database, an extension computing unit and an output device. The image capture captures a hand image. The feature capture connects to the image capture and captures at least one geometry feature of the hand image to produce at least one image matter-element model. The database has several recognizable models. The extension computing unit connects to the feature capture and the database. The extension computing unit can calculate a correlation between the image matter-element model and the recognizable models. The output device which connects to the extension computing unit compares the correlation with a threshold to output a recognition signal.

Description

201203131 VI. Description of the Invention: [Technical Field of the Invention] The present disclosure relates to a method of recognizing an identity, and more particularly to a method for recognizing an identity using a hand image.疋 [Prior Art] With the increasing popularity of Internet applications and the advent of the era of electronic trading, the traditional security and authentication methods have made it difficult to ensure the security of authentication. In addition to the user's need to memorize multiple sets of passwords, which may lead to errors, the use of passwords is more likely to cause fraud on corporate computer networks and the Internet. In addition, the use of plastic cards, smart cards or computer badges is not safe because they may be forged, stolen, lost, or unreadable. Therefore, the use of personal unique biometrics for identity identification is difficult to copy or steal, which can effectively solve the problem of security certification. Conventional hand recognition uses a ccd, a charge-coupled device or a camera to capture a hand image. However, • Traditional pick-up equipment is susceptible to identification due to factors such as palm print stains or injuries. In addition, traditional image capture equipment is also susceptible to external factors such as ambient light intensity, image background, angle, etc., resulting in poor identification results. SUMMARY OF THE INVENTION Therefore, one aspect of the present disclosure is to provide a system and method for identifying a hand image to overcome the above-mentioned problem of identifying results due to external factors. 201203131

According to the present disclosure-embodiment, a hand image recognition system includes an image capture device, a feature extraction unit, a recognition library, an extension calculation unit, and a wheel-out unit. The image capture device captures two hand images. The feature capture unit is coupled to the image capture device to capture at least the geometric features of the hand image, thereby generating at least an image matter element model. The second knowledge database has a plurality of identification matter metamodels. Extension computing unit The feature extraction unit and the identification database are connected, and the extension computing unit is used to calculate the image matter element model and identify an association degree between the matter element models. The output unit 连接 is connected to the extension computing unit, and outputs an identification signal according to the correlation degree and the result. According to another embodiment of the present disclosure, a method for recognizing an identity by using a hand image includes the following steps: capturing an image of a hand by using an image capture device; and capturing a hand image by using a feature extraction unit. At least one geometric feature, and generating at least one image matter element model; establishing at least one identification object element i using the extension unit true unit juice calculation image matter element model and identifying an association degree between the matter element models; The result of comparing the threshold values outputs an identification signal. Therefore, the identification system and method of the hand image of the present disclosure utilizes the geometric features of the hand image to identify the identity and reduce the influence of external factors such as angle, light, and environment on the recognition result. In addition, the extension computing unit of the present disclosure utilizes an extension identification algorithm to identify the identity, which not only establishes a simple and convenient model, but also has a strong model construction capability, can also save a large amount of memory space, and quickly classify. Repeated learning p can save time in identifying jobs. 201203131 [Embodiment] FIG. 1 is a flow chart showing a system and method for identifying a hand image according to an embodiment of the present disclosure. As shown in the figure, the identity recognition system and method of the hand image first uses the image capture device to capture the hand image 110, and performs image pre-processing on the hand image 120, and then uses the feature extraction unit to capture the geometric features of the hand image. 130, and using these image features to generate an image matter element model 140. At the same time, the embodiment further establishes an identification database 150, which includes a plurality of identified matter element models, and then uses the extension computing unit to calculate and calculate the correlation degree 160 between the image matter element model and the identification matter element model, and associates the relationship. The degree is compared with the preset threshold value 170. If the correlation degree is greater than the threshold value, the identification identity is displayed 180. Conversely, if the correlation degree is less than the threshold value, the person 190 is displayed. Figure 2 is a schematic diagram showing the identity recognition system and method of the hand image of Figure 1. The image capture device 210 is used to capture a user's hand image. The traditional hand recognition system uses a CCD or a camera to capture the image of the hand, and then captures the characteristics of the palm print and the palm shape. However, the conventional grasping device has the disadvantage that the palm print is susceptible to contamination or injury and affects the identification result. . In addition, traditional image capture devices are susceptible to ambient light and must be compensated for by light assist or light control. In this embodiment, the hand image is a specific infrared band signal that senses the heat radiation of the human body, and then converts the signal into an image image that can be recognized by the human vision without the problem of light interference, and the present embodiment 撷When the hand image is taken, the image capture operation is performed by the thermal image analyzer in a non-fixed manner, and the user can perform the hand image capture by simply placing the hand at a fixed distance. Table 1 below is a thermal image analyzer and a conventional CCD or scanner.

201203131 Feature Comparison Table. ^---- Capture Image Overwork Table 1~~~~~1------- Thermal Image Bifurcation--------Feature 1 --- . It is not affected by the ring thief sugar _ interference and affects the ship's capture. 2' f In the dark environment, you can still capture images. 3. Provide a non-contact capture feature. 4. There is no need to use a fixed cylinder to limit the user's finger opening process - just fix the distance of the palm. Conventional CCD or 1. The cylinder must be used to secure the palm of the hand and the user feels uncomfortable. Scanner 2. If the palm print is injured or dirty, there will be a result of identification failure. You must use the light to assist in obtaining the image. The sign, ...u ... π τ ~ is taken from the special part of the film family. Before this, the hand image must be pre-processed to capture the subsequent features of the feature acquisition unit. At this time, the hand image is gray, and the higher the gray level value, the higher the temperature, and vice versa. In order to leave only the hand grayscale value part of the hand image, the image processing unit 230 =· line image segmentation is used to separate the hand image from the background image to ensure that the part of the image and the remaining palm portion are grayscaled. When 0 (ie black), the feature is captured. This embodiment adopts the optimal threshold extraction technique proposed by Otsu. The main idea of the =tsu method is to find an optimal threshold value so that the sum of the two groups of variances is the smallest. Assume that the grayscale range of the image is 1~ especially, and the points of each gray P white are cumulatively, "2,..., and all the total points are #, then the probability of occurrence of gray Z• is:

LSI 7 201203131

N Now suppose that the pixel in the image is divided into two clusters (^ and C2 with a threshold A, (^ represents a cluster of pixels with a gray value range of [1,]; and c2 represents a range of gray values of |> +1, Z] pixel cluster. At this time, the probability distribution of the two clusters is w7, and the average of the two clusters is /X/, then:

K ^0=Σ,Ρι /=7 =w{k) (2) wi == /~wW i=K+J (3) μ〇= Σ Τί^ο =South\^k) (4) ^= ςιΆ= i=K+lw; uT —u{k) (5) The variation of the two clusters is &, <, the total number of mutations is <

Then: =ΥΜ~^)2 A (6) /=/ % i=k+I ? Pi wl (7) (6)= --w0a20(k)+\ (8)

Otsu believes that when a moxibustion value can be found so that the total number of variances is < the minimum, this A: is the optimal threshold. Next, the image pixel grayscale value is less than the A value is set to 0 (indicating black), and the greater than or equal to the threshold value does not change the grayscale value, as in equation (9), redistributing the grayscale value to the image can separate the background and The effect of the hand. (9) 201203131 Please continue to refer to Figure 2, after the completion of the split, then the background of the special leak (4) is connected with the image picker 21〇, and the feature is taken as 70 22〇 and the image object model. “The geometrical features of the image, which in turn produces the geometrical features of the hand, is one of the hottest items in the study of biometric systems. Since the geometrical features of the hand are minimally affected by light or the environment, it has nothing to do with skin color.

Only in relation to size, shape and length, the present embodiment utilizes the geometric features of the hand to perform identity recognition. FIG. 3 is a schematic diagram showing the geometric features of the hand image in the identity recognition system and method of the hand image of FIG. 1 . When extracting the geometric features of the hand, you must first detect the edge pixels of the hand. The size of the palm is represented by the number of pixels. When the edge is detected, the edge coordinates are recorded. After the coordinates are recorded, the fingertips and valleys can be found. The coordinate relationship is regularly variability. Using the regular relationship of coordinates, you can find the position between the fingers and the valley. In Figure 3, the points A to G are between the valleys and the points η to L are the fingertips. After finding the fingertip and the inter-valley coordinates, use the two-point coordinate distance formula to calculate the finger width and finger length. Use the relative distance between the fingertip and the valley to find the finger width. For example, the middle segment of the thumb is to take a coordinate point from the 1/2 distance of each point and use the two points to calculate the length. The width of the four fingers is also used to calculate the length at each of 1/3 and 2/3. The length of the finger is calculated by using the two valley points to calculate the midpoint. The distance between the midpoint and the fingertip point is calculated, and the length of the finger can be known. For example, the midpoints are found by the Α and Β, and the midpoint and the Η are calculated. The length is OK, and the other four fingers are also processed. 201203131 After the image pre-processing, only the hand area value is left in the hand image. At this time, record the number of pixels of all gray value values, and record the maximum and minimum gray level values in the 22nd domain. The second aspect of the present embodiment is as shown in Table 2 below. Special second-hand features 11 : Palm circumference Length of the thumb Index length of the middle finger Length of the ring finger _14 15

8 9 1/3 index finger width 1/3 middle finger width 18 1/3 small thumb finger window 1 mill 2/3 middle finger width _ 2/3 ring finger width & 2/3 small thumb finger width ^ 1 gray interval interval - --^ The second gray interval interval 铋---- The third gray-scale region 卩 the fourth gray-scale region, the fifth gray-scale region 卩 1/3 ring finger wide rabbit ------I_I a~~ day tSE After continuing to refer to Figure 2, after completing the above pre-operation, the extension calculation unit is used to identify the identity. In the identification of the user, the user must provide a hand image to create an identification database 24, so that the library 240 has a plurality of identification object models. In this embodiment, the feature of the image of the hand image of the eve is used as a training sample, and the eigenvalue of the training sample is used to establish an identification matter element model, and the identification of the matter element model can be established, and the database 240 can be verified in the future. In time, the identity verification unit 250 can be used to calculate the image matter element model and identify the degree of association between the matter element models, so that the identity verification can be performed. On the other hand, the output unit 2 shame extension unit connection 250 outputs an identification signal based on the correlation degree and the threshold value 201203131. The following is a basic introduction to the extension theory: The extension theory was published by Professor Cai Wen, a mainland scholar, in 1983. The two pillars of extenics are matter-element theory and extension set theory, which can be described by using matter-element models. A comprehensive analysis of quantitative and qualitative changes. Therefore, both the quantity and the quality can be studied simultaneously, and the user can get a complete message about the authenticity of the system characteristics. First, the definition method of the matter element is introduced. Let the name of the object i? be 7V, and the quantity of the feature C is F, then the basic element or matter element describing the thing is (10): R = (N, C, V) (10) If a matter element has multiple characteristics, it is called a multi-dimensional matter element. Let the things have w characteristics of C7, C2, ..., C„, and their corresponding magnitudes are respectively, ..., then the expression of the matter element is (11): 'NC, v, R = * c2 • V2 >=- r2 > c„ ln K. Λ.

The definition of the extension set: set the domain, if any element w and w @, then there is a real number A: (8) @ (- 〇〇, 〇〇) corresponding to it, then the extension set is (12) ^ = {(u,y)\u 6 U,y = K(u)e f-〇〇,〇〇;} (12) and less = You〇) is a correlation function of 2, ruler (9) is the degree of association of w, The range is -〇〇 to 〇〇, and its extension set 2 can be expressed as (13) in the U universe: A = A+ u J0u A~ (13) where A+ = { (u , y^ug U , y = K( x) > 0 } (14) 5 3 11 201203131 J 〇= f (μ, y^uau , y ~ κ (x) = o } A = , y'u & u , y = K( x) <〇} 1 (14), the positive domain called j' is called the negative domain of 2, and ^ is called the zero bound of J. The correlation function has different forms in application. For =<~于, and <^, do, for @1 and no public endpoint, the elementary correlation function is represented as (8): K(x) = D(x,Xo,X) (15)

The point is: the distance from the interval A is defined as (9) Ρ(χ>^〇)= X — a + b ~Ύ~

(16) In addition to considering the positional relationship between points and intervals in the extension set, the current item considers the positional relationship of one point to two intervals. Let ^^<4>,Z=<Ci^ and be 'with point X about the value of the non-/, D), D{x, X〇, X) = \p(^x^x) ~ P{x,X〇) xiX„ l'1 x eX〇(17) mi m

The number of stuns is not related to the degree of association, when the heart Λ Γ Γ! 0 degree you) $0 is called X does not belong to the degree of &; When it is changed, 4 has the opportunity to become this ratio: ίτ sorts out the traditional neural network and the extension identification algorithm. The traditional neural network performance algorithm needs to be trained. The work time is 12 201203131 2. Update the key value and the inter-value value in an iterative manner, and the calculation amount is large and consumes computer resources.

The library must be re-learned. The applicable fields are wide, and the model construction ability is strong. 2 'Building a model is simple and convenient, saving memory space. 3. It can save a lot of modeling space and classify quickly. The library does not need to be re-learned. The extension identification algorithm, ------two _^ bamboo 厍 does not require Dong Xin learning. In the following, the body image of the experiment 7 of the present disclosure will be disclosed (4). It should be understood that the requirements mentioned in the above embodiments will not be described in the following descriptions, and those who have been further defined are summarized, and the materials are «reported, only need to be applied. (4) The hand image of the self-finance is closed. Part of the image, a total of hand image, using this hand shadow to create a recognition object model, so there will be 2q identification object models stored in the identification database. The hand mail then personally takes 1G of images that are not in the identification database, so that another 200 hand images are generated. For the identity type, if the image is created as an image element model ηΑ Μ 20 20, the type of image and the image of the model are calculated ft 13 (19) 201203131 The correlation degree of the matter element model . -ρ(Κ,νϋ) & P(v“,vy) piv.y^-piv.y,) where P(Vti,Vpi) yti- K- + by Ύ~ oi + K 2 by - aij ~~ T~_Kzi 2 ' 1,2,...,20; j = 1,2,...,20

The weights of the identity verification are set according to the values of the correlation functions, and ..., and the weight values of the present embodiment are all set to 1/20. 20

j=J Reuse equation (20) to calculate the degree of association of each verification identity. 20 into i = work % & i = 1, 2" _" 2 〇 (20) j = l

Finally, the identity of the hand image is confirmed, and the recognition result is determined according to the degree of relevance. The person with the highest degree of association and greater than the preset threshold value can confirm the identity to which the hand image belongs. If the degree of association is greater than the threshold value, the identity is displayed. Otherwise, it is displayed that there is no such person. After the experimental test, the other 200 hand images and the identification database were identified. The results are shown in the following table IV: Table 4: Hand image identification success identification failure identification rate training 200 200 0 100% identification 200 198 2 99% 14 201203131 Prove that both the training and the identification image, the identification::: The image identification system and method of the image has a very high recognition algorithm experiment: two = is the implementation and practice Knowing hand

201203131 [Simultaneous Description of the Drawings] FIG. 1 is a flow chart showing a system and method for identifying a hand image of an embodiment of the present disclosure; FIG. 2 is a diagram showing a system for identifying an image of a hand image of FIG. Unexplained, FIG. 3 is a schematic diagram showing the geometric features of the hand image in the identity recognition system and method of the hand image of FIG. [Main component symbol description] 110 to 190: Step 210: Image capture device 220: Feature extraction unit 230: Image processing unit 240: Identification database 250: Extension calculation unit 260: Output unit A to L: Point

Claims (1)

201203131 VII. Applying for patents: 1. The identification system of the hand image, including: an image on the device to fetch a hand image. - Feature Na unit, Lai Jingyu _ (4) at least the geometry of the connected image The feature further generates at least one image-identification library, and has a plurality of identification object models, and - the extension computing shirt 'sorts the hash and the identification library connection' (four) (four) can calculate the image object And a correlation unit between the meta-model and the identification matter element; and an output unit connected to the extension calculation unit and outputting an identification signal according to the comparison result of the correlation degree with a threshold value. 2. The identity recognition system for a hand image according to claim 1, wherein the geometric feature is a grayscale distribution statistic of the hand image. 〃 3. The identity recognition system for a hand image according to claim 1, wherein the geometric feature is a perimeter of the hand image. The identification of the hand image of the hand image as claimed in the item ,, the geometric feature of the basin is the length of the hand image. The system identification of the hand image as claimed in claim 1 is the width of the image of the handcuff. </ RTI> [S] 17 201203131 6 The identification system of the hand image according to claim 1, further comprising: an image processing unit for separating the hand image and its background. 7. The identity recognition system for a hand image according to claim 1, wherein the image capture device is a thermal image analyzer. 8. A method for identifying an identity by means of a hand image, comprising: • capturing an image of a hand using an image capture device; capturing at least one geometric feature of the hand image using a feature capture unit and generating at least An image matter element model; establishing at least one identification matter element model; calculating an association degree between the image matter element model and the identification matter element model by using an extension calculation unit; and comparing the correlation degree with a threshold value according to the correlation degree The result outputs an identification signal. 9. The method for recognizing an identity by hand image according to claim 8, further comprising: reallocating the grayscale value to the hand image by using an Otsu's method to separate the hand image and its background. 10. The method of identifying an identity by a hand image as claimed in claim 8, wherein the image capture device senses the infrared band signal in a non-contact manner.