KR20110068688A - A three-dimensional figure measuring system for use in three dimension figures thing production - Google Patents

Abstract

PURPOSE: A three-dimensional figure measuring system for manufacturing three-dimensional figures is provided to offer more exact recognition rate than the existing measuring system. CONSTITUTION: A three-dimensional figure measuring system for manufacturing three-dimensional figures comprises the steps of: subdividing head data using a marker position; separating a cross section from points on the surface of the collected head data and abstracting a contour line; checking noise tip identification of an interesting area from a contour map obtained from the abstracted contour line, and abstracting the cross section contour line; compensating a centroid distance vector uniformly using a function abstracting and contour line searching application; and making a performance analysis and defining match points.

Description

A three-dimensional figure measuring system for use in three dimension figures thing production}

The present invention relates to a three-dimensional shape measurement system for manufacturing a three-dimensional shape, and more particularly to a three-dimensional shape measurement method for extracting various data for manufacturing a three-dimensional shape from the surface shape data of the three-dimensional shape. I would suggest. This method relies on the three-dimensional shape signature extracted from the contour data. This eliminates the need for registration orientation. The shape signature is created by calculating the centroid distance of the translation and rotation boundary points. Shape signing is done using a signature and authentication function as a vector for the shape of the region of interest.

As the most common and representative example of the three-dimensional shape, biometrics refers to the recognition through the physical characteristics and behavioral characteristics of ordinary people. In the past, people were identified by pictures, passports, and ID cards. Similarly, entering a restricted location would require knowing a password or password. Biometrics skips this verification and certification process and lets you just show who you are and how you behave. Who that person is is a person's physical or genetic traits, and it means his or her own distinctive behavioral patterns or skills. Fingerprints, faces, hands, palms, ears, irises, and retinas are included in this physiological aspect. Beware of behavioral biometric features such as voices, walking shapes, handwriting, signatures, and keystroke pattern recognition.

As hardware and software technologies in biometric systems continue to develop, they become competitive in terms of price in the mass market. In addition, advances in computing power, networks, and database systems have made biometric systems much more widely available geographically and on networks. Today, many public institutions and companies are using biometric systems. Manufacturers have increasingly supplied biometric systems with computer equipment. Many manufacturers offer biometric sensors and their devices, including keyboards and laptops with fingerprint sensors in biometric system options.

As security becomes more important today, biometrics make it easier to worry about losing or stealing multiple credentials, such as cards and passwords. This quick and easy recognition device is more secure than keys, cards and passwords and can be prevented from being lost or forgotten. Such devices are unlikely to change relatively over time.

But while biometric devices have many advantages, it should not be overlooked that they also have vulnerabilities and limitations. Considerations for the construction of biometric devices include the identification of biometric characteristics and the level of originality, how easy a user can register and re-register, and template and database management. And a biometric device can never be a flawless security device. Many studies have been conducted to increase security, convenience, and reliability through biometric devices.

Accordingly, the present invention has been made to compensate for the shortcomings of the above-described biometric system, that is, the three-dimensional shape measuring system, and the three-dimensional shape for the production of three-dimensional shape using the new technology recognition method using the surface data developed by the present inventors. The challenge is to provide a measurement system.

The three-dimensional shape measuring system for producing a three-dimensional object of the present invention solved the above problem is the step of subdividing the head data using the marker position; Extracting a contour by separating cross sections from points on the surface of the collected head data; Identifying a nose tip identification of a region of interest from a contour map obtained from the extracted contour, and extracting a cross-section contour; After extracting the contour, correcting the centroid distance vector using boundary spline interpolation using a function extraction contour search application; And correcting the centroid distance vector, and performing a performance analysis to define a match score.

Here, the extraction of the cross section contour is characterized by satisfying the following Equations 3 to 5.

(Equation 4)

Calculate as above in area A.

Centroid is set to C = (Cx, Cy)

(Eq. 5)

(Equation 6)

In this case, the centroid distance vector using the function extraction contour search application is defined according to Equation 7 below.

(Eq. 7)

(Where C _j = (C _jx , C _jy ) is the centroid of the J _th contour, and (X _ji, Y _ji ) is the point of the boundary.)

Here, the collection of the head data is characterized by using a Radon transform that represents the model image of the object as a line from 0 to 360 degrees.

The three-dimensional shape measuring system for manufacturing a three-dimensional object according to the present invention is an improvement of the conventional three-dimensional shape measuring system. When the three-dimensional shape measuring system of the present invention is applied, the drafting error rate is 4.563%, which is higher than that of the conventional measuring system. Providing accurate recognition rate, it is possible not only to construct more accurate data for manufacturing a three-dimensional shape using this, but also when applied to the living body, there is an effect that can be directly applied to various security systems.

Hereinafter, the present invention will be described in more detail.

1, shown in accordance with one embodiment of the present invention, illustrates the major components of a typical stereoscopic measurement system.

2 illustrates an example of a ROC graph model.

3 illustrates an example of several biometric features.

4 shows four different postures according to an apparent database.

FIG. 5 shows a landmark marker (a) in a three-dimensional face and a three-dimensional shape (b) from which markers and information have been removed.

6 is a cross section contour extracted from a three-dimensional shape from which markers and information are removed.

7 shows a demonstrating nose tip procedure.

8 shows four selected contour function extractions.

9 is a graph illustrating data for setting different threshold values between an actual user and a fraudster.

10 is a graph showing the work characteristic curve (a) and the decision flow curve (b) of the prisoner.

11 is a diagram flow diagram of the present invention.

12 illustrates a change in radon according to an embodiment of the present invention.

Figure 13 shows the collected hand image according to an embodiment of the present invention.

14 illustrates grayscale binary conversion according to an embodiment of the present invention.

Figure 15 shows the state after successfully clearing the wrist portion in accordance with an embodiment of the present invention.

16 illustrates an example for calculating the centroid distance after erasing the wrist part according to an embodiment of the present invention.

17 illustrates an example of a compressed position of an invariant function according to an embodiment of the present invention.

18 illustrates another grain boundary value distribution (a) and an ROD curve (b) of FAR and FRR according to an embodiment of the present invention.

19 illustrates a flow diagram according to an embodiment of the present invention.

20 illustrates X, Y, and Z axes associated with a foot in accordance with one embodiment of the present invention.

FIG. 21 is a graph illustrating general force measuring plate data using three coordinate axes according to an embodiment of the present invention.

FIG. 22 illustrates gait data along the Z-axis before the wavelet removes noise according to an embodiment of the present invention.

FIG. 23 is a diagram illustrating gait data along the Z axis after the wavelet is noise canceled according to an embodiment of the present invention. FIG.

24 is a graph illustrating three times the force measurement plate data of the X axis of the object 1 according to an embodiment of the present invention.

FIG. 25 illustrates a ten store histogram corresponding to FIG. 24.

Figure 26 illustrates a procedural step of finding histogram similarities according to an embodiment of the present invention.

The present invention comprises the steps of subdividing the head data using the marker position with a three-dimensional shape measurement system; Extracting a contour by separating cross sections from points on the surface of the collected head data; Identifying a nose tip identification of a region of interest from a contour map obtained from the extracted contour, and extracting a cross-section contour; After extracting the contour, correcting the centroid distance vector using boundary spline interpolation using a function extraction contour search application; And correcting the centroid distance vector, and performing a performance analysis to define a match score.

Here, the extraction of the cross section contours satisfies Equations 3 to 5 below.

(Equation 4)

Calculate as above in area A.

Centroid is set to C = (Cx, Cy)

(Eq. 5)

(Equation 6)

Here, the centroid distance vector using the function extraction contour search application is defined according to Equation 6 below.

(Eq. 7)

(Where C _j = (C _jx , C _jy ) is the centroid of the J _th contour, and (X _ji, Y _ji ) is the point of the boundary.)

Here, the collection of the head data uses a radon transform that represents the model image of the object as a line from 0 to 360 degrees.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings. In particular, in the following description, it is exemplified that the three-dimensional shape will be illustrated as a living body having the most general, representative and complicated shape, and the measurement of the three-dimensional shape will be described by substituting the biometrics. However, the following description does not limit the present invention.

The importance of the present invention is to introduce some of the techniques of new biometric devices. Various biometric factors such as face, hand shape, and forced profile are studied and function extraction methods are also developed and evaluated.

The present invention consists of five parts: biometric systems, forms, and performance certifications. First, there is a brief introduction to biometrics. There is a description of important parts of a normal biometric system. The terminology for each performance analysis is explained. Physiological and behavioral characteristics were also briefly introduced. The pros and cons of biometric systems have also been described.

 Second, facial recognition method using 3D surface shape data is introduced. The 1D surface geometry process is introduced and the research results from 100 subjects are introduced.

Third, 18 objects for hand recognition consist of 136 low resolution images. It shows a good case of moving invariant properties through radon.

Fourth, there is a description of the biometric device according to the ground reaction force seen when a person climbs on a pressure sensitive mate. Force profiles are used for authentication, and histogram-based methods are used for recognition decisions.

Fifth, there are guidelines for future research and a summary of the results and methods of this study.

First, an overview of biometric systems

Biometrics, derived from the term 'biometric technology', has recently been expanded to include a study of methods for biometrics, meaning that the meaning of the term depends on one or more of the physical or behavioral characteristics of human nature. Set up the function with a biometric system and obtain several personal data in comparison with a previously acquired registration database. There are several key aspects that contribute to the development of the biometric system, robustness and specificity. Robustness refers to the consistency of biometrics for automated measurements. Robustness is a preservation and stability function of a biometric system. Specificity refers to the distinction of a person's unique biological characteristics from others in a group of users. For example, fingerprints are easier to distinguish than the shape of a hand.

Biological function data is usually performed in two situations. One of them is the 'closed set' operation from which the data is known. The other may come from a user not known as the "public set." Or there is a situation where the system has no choice among the 'options' above. The biometric system can also operate on two different modes of authentication (or verification) and identification. In authentication mode, a user authentication called "one-to-one mapping" is required, in which case the user asserts his or her identity by presenting one or more of a biometric function and system to accept or reject his claim. have. The identification mode system identifies one from a set of known users by performing one of the 'many comparisons'. In general, a configuration recognition algorithm, in which a biometric system is similar or which accepts or rejects authentication of a user according to a matching score, is as follows.

Biometric systems identify various physical or behavioral characteristics through fingerprints, faces, hands, irises, retinas, signatures, gait, palm prints, voice, ears, hand veins, odors, and DNA information. In literature this is called a characteristic, identifier or form. Despite their limitations, biometric systems are not easily stolen than conventional security measures. In addition, the biometric system enhances user convenience by memorizing and reducing the design and password as needed.

Advances in digital signal processing have made fingerprint, voice, hand and face data suitable for automated systems. Faster computer availability and improved detection technology have been developed to make it more suitable for biometric data in iris, mandala, gait, etc., with significant advances in computer vision and statistical pattern recognition. 3D modeling and graphics have also paved the way for processing 3-D biometric data such as various images of hands, faces and ears.

 [Types of Biometric Characteristics]

Biometric properties include genetic physiological and behavioral properties. A brief description of each feature follows.

 Genetic characteristics

This includes hair color and eye color inherited from parents. In the case of facial skeleton, it is very difficult to change, so it is good to use as a differentiated characteristic.

Physiological characteristics

This is the answer to the question 'Who am I?' That is, there are patterns such as fingerprints, hand shapes, irises, and retinas that a person has.

Behavioral Characteristics

This character is a learned or trained character. These characteristics include voice, sine, gait and so on. According to the theory, these characteristics can change or relearn. But in the general case of adulthood, these traits are difficult to change unless there is special and rigorous training.

All biometrics include genetic, physiological and behavioral characteristics. For example, there is a behavioral characteristic when a person touches a finger or face to a biometric device. Basically, popular biometrics uses a biometric system based on genetics.

Components of a typical biometric system will be described with reference to FIG. 1.

1 shows the general flow of a biometric system. The system has the following components.

<Data acquisition>

In this process, biometrics are shown. It converts the biometric digital search and capture process and the resulting data into signal processing. If data is captured at a remote location, it is compressed or coded before being converted.

<Transport channel>

This process means a way of communicating between fundamental functional constructs. Some systems are self-contained or have a transmission channel in the unit. Other systems choose to store and transfer data collected from multiple locations in a central data store. Systems that can interfere with the transmission channel include the LAN, the private Internet or the general Internet.

<Signal Processing>

In this section, the unorganized biometric data will be matched against a pre-captured template. Fine-grained samples and materials are used to create new templates. Segmentation means the separation of relevant biometric data from the background. Newly created templates based on ID or ID are compared with one or more related templates according to matching algorithms. The result of the matching algorithm is the match score and indicates how similar the templates are.

<Decision method>

This is the final step in signal processing that determines whether or not a match is made. Normally, a new match will occur if you use a threshold you've experienced or if your score drops. The decision method introduced is an example of a general decision method. In other cases, it is determined by the real factor according to the real factor requirement.

[Performance Assessment of Biometric Systems]

Performance evaluation of the recognition system is usually scored according to the 'successful recognition rate' or 'correct recognition rate' (Equation 1).

(Equation 1)

 CRR = the number of people correctly recognized / total people * 100%

Biometrics or discrimination is the discovery of a trade-off between two types of errors. The first error is when the real user is rejected for a system problem (False Rejection), does not match or is not detected. In the second case, a fraudster accepts a real user. These two errors can act as a factor in changing the threshold. More secure security systems avoid low FAs and high FRs. In general, completed tradeoffs that gently cross the operating point are used to identify biometric performance. The false acceptance rate / False Acceptance Rate and False Rejection Rate are defined as follows (Equations 2 and 3).

(Equation 2)

FAR = number of authorized cheats / total cheats * 100%

(Equation 3)

FRR = Number of actual users rejected / total actual users * 100%

Performance can be viewed as a Receiver Operating Characteristics plot, and FRR is a factor that changes the threshold for FAR. Looking at the ROC plot in Figure 1.2, the lower left corner of the plot is the preferred value and the area that minimizes the two error components. However, the ROC plot of a well-performing system is mostly met in the lower left corner, which hinders a clear view of the competition system. Recently, a variety of ROC plots and a modified Error Error Offset plot were used, which used a normal deviation scale. This affected the movement of the curve in the lower left corner, creating a more straightforward curve with higher performance.

Although a complete DET or ROC curve is needed to account for system error tradeoffs, it is suitable for reporting performance evaluations in single digits. Often the Equal Error Rate (EER), which appears on the ROC curve when FAR and FRR are equal, uses a single digit.

Except for FRR and FAR, biometric systems have two more errors. One is Failure to Capture (FTC), which is when the system fails to capture the characteristics of the sample. This can happen when something is wrong with your digital camera and fails to capture clear images. Another error is Failure to Enroll (FTE), which represents the percentage of users who cannot successfully enter the system. User training is necessary to obtain good biometric data (see Figure 2).

[A brief description of the different biometrics detailed principles (see Figure 3)]

Here's a quick introduction to commonly used biometrics.

 <Face>

Faces are the most biological features people use to recognize each other. The most commonly used facial recognition methods are the spatial analysis of face contours such as eyes, eyebrows, nose, lips, and chin, and the overall analysis of face images based on the combined weight of basic facial figures. In popular systems there are several suggestions for obtaining face images. This includes a fixed simple background and controlled lighting. In this system, it is difficult to match when face images are captured in different lighting conditions or in two different eyes. To make a facial recognition system better, you need to be able to automatically capture the image you want on your face and to recognize it from a common viewpoint.

<Fingerprint>

The accuracy of fingerprint recognition is known to be very high. <16, 17> Fingerprints are patterns of curves that appear on the surface of the fingertips and are produced during the first seven months of fetal formation. As is widely known, the fingerprints of identical twins are different as well as the fingerprints of each ten fingers of an individual. Currently used fingerprint recognition systems are reasonably priced and accurate enough to be used in criminology. Transmitting multiple fingerprints of an individual to the system can provide the information needed for wider identification. The only problem is that the fingerprint recognition system requires a lot of information and resources, especially in identification mode.

<Hand shape>

The hand recognition system is based on individual hand types, palm size, finger length and width. General hand recognition systems are used in many places around the world. The technology in this system is very simple and easy to use and inexpensive when compared to others. But hand-based analysis is very difficult to find, and you can't use this system to deal with a large number of people. Also, children who are growing up cannot use this information because it changes. In addition, each person's jewelry wearing or dexterity limits interfere with accurate hand shape information.

<Palm print>

This is a pattern of curves and hills in the palm of a person, just like a print. The print on the palm is more accurate than the print because it is more space than the print. The palms of people have different shapes of palms and wrinkles, which are cheap in terms of price because they can be captured by low-resolution scanners. Thus, capturing all these hand features with a high resolution scanner creates a highly accurate biometric system.

<Iris>

The iris is an annular area of the eye surrounded by pupils and sclera. The characteristics and appearance of the iris form in the fetus and within two years of life these points stabilize. Complex iris characteristics can be very helpful in recognizing people because they vary from person to person. Recently, the iris-based recognition system is very satisfactory in terms of accuracy and speed, and acquires iris information that can be used in a large scale recognition system. Each iris is different, and identical irises are different. Iris-based recognition systems have very low FAR compared to other biometric systems, and FRR is high.

<Voice>

Negative is a combination of physiological and behavioral characteristics. The physiological characteristics of an individual's voice are all that is used to make sounds such as saints, mouth, nasal cavity, and lips. These physiological traits are immutable, but behavioral traits change with age and can be affected by a cold, or emotional state, in medical terms. Voice doesn't have much differentiation and is not suitable for use in a wide range. The text-based speech recognition system allows a predetermined sentence to be recognized when speaking aloud. Text-based speech recognition systems, on the other hand, are intended to prevent fraudsters by analyzing and recognizing what speakers are saying, but much harder to design. The disadvantage of speech recognition systems is that they are exposed to noise. It is suitable for identification on the phone, but voice signals are of poor quality in communication channels.

<Walking>

It refers to the shape of a person's walking and is one of the biological characteristics that is difficult to recognize at a time. This is therefore very suitable for surveillance systems. Most gait recognition algorithms analyze human silhouette according to the space time model. Therefore, the more diverse and good human models, the higher the functionality of the gait recognition system. Some algorithms use optical flows to read individual steps by substituting moving points into the body <25>. This system can be used to identify individuals over time. However, gait can be affected by the type of shoe, the type of clothing, the condition of the legs, and the ground.

<Ear>

Ears are often compared to facial recognition. Ears have several advantages over faces. The measurement range is narrowed, the color uniformity is increased, and it is less affected by emotion or facial expression than the face. In facial recognition systems, results may vary depending on lighting and position. This problem is the same as the ear recognition system, but it can be an advantage because the ear is a smaller image than the face.

[Benefits of biometric systems]

There is a reason that biometrics are becoming popular.

<Convenient Recognition>

It's much faster and simpler than using a card or key password, and the recognition process itself is smooth. Bionic technology allows you to identify yourself using your physical characteristics without worrying about losing or forgetting. Many users remain relatively unchanged over time.

Need for Strong Security Awareness

Passwords and passwords are easy to be stolen. Biometrics reduce this risk. Modern society demands stronger security for computers and physical access, and biometric systems are much more secure and less susceptible to being stolen than cards and passwords.

<Price lowering>

In the last few years, advances in hardware and software have made biometrics prices competitively competitive in the mass market. In addition, advances in computing power, networking, and database systems have enabled biometric systems to be used in larger areas and networks.

<High adoption of public institutions and enterprises>

Many public institutions and corporations today use biometric systems. After September 11, public safety and security have emerged, raising interest in biometrics. Manufacturers have increased the supply of computer devices with their systems. Many companies have included biometric sensors and customized devices in their system options. For example, a fingerprint sensor is installed directly on a keyboard, mouse, or laptop.

[Biology of Biometric System]

While this system has many advantages for both the public and government, problems can arise when used in a wider market. Such problems are as follows.

<Obstructions in acquired data>

The faulty data is a result of sensor problems or environmental factors. For example, noisy data can be a scarred fingerprint or a cold voice. This can result in the end user being denied access because the impaired biometric data may not match the appropriate template.

Intraclass change

The main cause of intraclass changes is when an individual reacts inaccurately with a sensor, or when biological characteristics change over time. These changes can be resolved by comparing the various templates of all users and updating the templates <30>. These changes are more likely to occur in behavioral properties that may exhibit different behaviors over time than physiological properties. For example, a sample included in a voice sample and a voice used for recognition may have different results according to an individual stress index.

Intra Class Similarities

This is a property that is commonly overlapping between individuals <14>. The more people registered in the identification system, the greater the interclass similarity between individuals, which can lead to more system errors.

<Not all-round>

The biometric system may not be able to extract available data from the user. In the case of a fingerprint recognition system, if the lines of the fingerprint are not clear, it may not match the personal data. Using one biometric feature increases FTE (Failure to enroll).

Information Interoperability Issues

Most biometric systems consider biometric data to be compared using the same sensor, which limits the comparison and analysis of biometric data between different sensors. For example, speech recognition is difficult to compare in two different technologies, electret and carbon-button. Similarly, fingerprints can be changed by several sensor technologies, including changes in sensor technology, image resolution, measurement location, and adverse effects. Whenever the sensor changes, the cost and time used to re-register again is enormous and causes a big problem for the user's convenience. Therefore, a technique that enables interoperability between feature extraction and matching algorithms to interoperate with other devices becomes more necessary over time.

Fraud Attack

Fraud is the act of deliberately misleading an individual's biometrics, calling for cognitive errors, or stealing another's identity by creating fake biometrics. This fraudulent behavior uses signatures and voices, which are behavioral characteristics. There are several ways to report this fraud. When using physical characteristics such as fingerprint and iris, a real-time detector scan can be used.

[New technology recognition method using surface data]

Faces are the easiest way to identify individuals. People can distinguish themselves from others without much effort, but facial recognition systems are much harder than they think. Problems include raising prices when there are many images. The change in the inter and intra subjects is in the change in the face image. Similarities between individuals are responsible for changing subjects, and intra subjects are influenced by a person's age, head pose, light, and the presence of other objects or people.

Automatic facial recognition has been widely used in the last few decades. Image processing pattern recognition, computer vision, networks, artificial intelligence, biometrics, etc., have all been active. Many researchers have worked through a number of studies to make machines and computers recognize people as if they knew people.

Facial recognition can be used commercially, in security, and in criminology. It is used in many genres, including automated public surveillance systems, access control, driver's licenses, credit card issuance, ATM rights, and HCI.

The face recognition process uses stored face data to identify an individual in a given image. The face recognition system recognizes by comparing and analyzing the image data collected in the gallery image as an image called a test image. The best match on the database is identified by the image on the system. There are two methods of facial recognition. When verifying, based on the characteristics of the individual, the system compares the user according to a predetermined algorithm or rules of outcome and decides whether to accept or reject it. In the case of verification, the method of determining who scores the most similarly among all the registrants is determined.

Until the early nineties, most of the work was done on two-dimensional images. Many new technologies have emerged, increasing recognition rates. The rapid assessment of facial recognition technology consists of the new system estimates FERET, FRVT 2000, FRVT 2002, XM2VTS protocols and existing software packages FaceIt, FaceVACS, FaceSnap Recorder, Cognitec, Eyematic, Viisage and Identix.

According to Brunelli and Poggio, face recognition is divided into feature-based and template-based contact methods. The first approach uses graph matching and the AAM (Active appearance model) to find features that are distinct from common features. On the other hand, the template-based method uses Eigenface and Fisherface to recognize the face by finding the association between the face and the face data template.

Most two-dimensional facial recognition systems are sensitive to lighting or pose changes. These problems arise because the two-dimensional image of the face has incomplete information. On the other hand, a three-dimensional scan can read the subject's face perfectly. It is well known that 3D images have much higher success rates than face counterparts. The rapid development of 3D image technology raises a lot of interest in 3D face recognition.

[3D Facial Recognition and Certification Status]

Bowyer introduced a survey of three-dimensional face recognition technology in his paper. Some techniques are derived from such things as Principal component analysis, a two-dimensional face recognition technique. Other techniques are unique to 3D facial recognition technology, such as surface matching, profile matching, and equidistant variation.

Cartoux et al. Modeled images of the left and right symmetry by analyzing images based on principal curvature. This model standardized the poses. Researchers say that the method of using a symmetric model with a face surface match has 100 sets of recognition rates.

In <52>, the size of each image has an attribute value as a metric size based on the segmented curvature of the image. Thus each face becomes a point in the functional space and the matching between the nearest neighbors ends.

Nagamine et al. [53] standardized facial pose using five function points. Instead of using the whole, it is possible to reduce the cost problem by allowing recognition using various matching curves and profiles through face data. The experiment is given 16 subjects and 10 images for each subject. The best recognition rate occurs when using a vertical profile curve and a curve passing through the center of the face.

Achermann et al. Use two-dimensional facial recognition techniques with eigenface and HMM models with various images. They showed a 100-perset recognition rate using a two-dimensional facial recognition algorithm in a database based on 24 people and 10 images per person. The convex regions of the various images are subdivided by their meaning and Gaussian curvature, and an extended Gaussian image (EGI) is created for each convex region. The recognition process is based on the correlation of EGI. The experiments were reported to have a 100 percent recognition rate if they had 37 images in the Canadian National Research Association's various image datasets. Principal component analysis used Hesher and many others in eigenvectors and image sizes. Six different facial expressions were used on 37 subjects.

Medioni and Waupotitsch (60) studied three-dimensional features captured by passive stereo sensors. Face recognition uses an ICP (iterative closest point) to match the face appearance to realize a three-dimensional shape using a stereo passive sensor. EERs for 100 subjects are reported at 2 percent and 7 different poses are provided each.

Moreno et al. [61] generate segmented region feature vectors using segmented three-dimensional face recognition based on Gaussian curvature. They represented the 420 facial meshes of 60 people and sampled each person's different facial expressions and poses. This frontal based subset had a 78 percent recognition rate.

Pan et al. [62] achieved three-dimensional face recognition using the Hausdorff approach and PAC-based approach. With images from the M2VTS database, they found three to five percent EERs with the Hausdorff approach and five to seven percent with the PCA-based approach.

Lu et al. (65) reported on three-dimensional face recognition through an ICP-based approach. In this experiment, images of 18 people succeeded in achieving 97 percent recognition rate, with different experimental images, altered poses and facial expressions.

Bronstein et al. Reported the three-dimensional face recognition in relation to facial expressions. The idea is to transform three-dimensional face data into what is called an eigenform, which can distinguish identical twins by transforming modeled images.

Lee et al. [69] presented a method for three-dimensional face recognition using eight feature points of a face. Using the support vector machine (SVM) to recognize 100 people, the highest recognition rate reached 96 percent. They used cyberware sensors to register images and captured experimental images with Genex sensors.

Russ et al. [71] studied the representation of various images of three-dimensional face data through the Hausdorff approach matching. Similar to the ICP, the interactive registration method uses an algorithm for finding experimental images from gallery images. In many ways, research has been conducted to reduce the complexity of time and place. In FRGC version 1, the results were shown: one experimental image per person was used rather than all experimental images. The highest recognition rate in practice was 98.5 percent, with 93.5 percent confirmation with a 0.1 percent FAR.

Chang et al. [68] introduced a three-dimensional face recognition approach. This combined the case where the overlapping areas around the nose were matched using ICP and the results of several three-dimensional matches. In the performance assessment, this study used 4000 images of 400 people, using the FRGC version 2 data set. In this experiment, if one image of the neutral expression was used in the gallery images, the other images were used for the experiment and 92 percent recognition rate was reported.

It is an object of the present invention to propose a new method for three-dimensional face authentication from surface shape data. This method relies on the shape signature of the face extracted from the contour data. This eliminates the need for registration orientation. The shape signature is created by calculating the centroid distance of the translation and rotation boundary points. Shape signing is done using a signature and authentication function as a vector for the shape of the region of interest. In this proposal, the scheme equality error rate was achieved of 4.563%.

Surface shape data is scan data including four postures that capture head data using a body scanner on a subset of all the bodies used in the present invention. Surface images are images of surfaces in some specific area of the surface, such as ear or nostrils. Facial hair is also of great interest. So the head is covered with a plastic cover during the data. A total of 100 subjects were used in this study. 4 shows an example of four postures per person. Data parts are distinguished using marker information. One gallery image of each subject is used and the other three are used as probes. Subsequent sections will discuss the proposed method.

<Method of the present invention>

The method of extracting the function separates the head data from the surface shape data of the whole body at the starting point. Marker and marker coordinate protocol information is used for this purpose. After separating the head data, the separated cross section is selected and identified along the z-axis for the contour line, its nest nose tip, and the area of return (ROI). Centroid distance is calculated from the contours selected for the shape signature. By removing the effect of the change that caused the data in the Discrete Fourier transform on the contours of the surface, we can obtain the final functional vector. All of this feature extraction is done in Matlab.

<Segmentation of Head Data Using Marker Position>

Surface data can be read from a triangular surface mesh of the form "-ply". The surface is colored according to the height of the point. Correct the face color of the triangle. Provides color information to identify the marker position. 5 (a) shows the marker position on the face surface. Green colored marker pictures are numbered on the face surface and marker 7 color coded data can also be used to separate the remaining head data. 5B shows the separated head data.

<Contour Extraction>

The z-axis cross section was separated from the points on the surface. Simple indexing works. FIG. 6 shows the contour extracted from FIG. 5.

<Select a region of interest: Nose tip identification>

This way, you don't need to register the data like you would a normal orientation. But for it to work with cross-sections, there must be a one-to-one correspondence between the two data sets. More precisely, it does not match the area around the nose and the area around the eyes in the cross-gallery image. Correct contour selection for authentication requires a common area of interest (ROI). Using knowledge of everyday life, it shows the most distinctive features in the forehead of the area between the nose and the lower part of the eyebrows and includes the first contour identification end. And for simplicity, choose one from more than 10 contour maps. Therefore 21 contours were used for the full feature extraction.

It is assumed that C = (Cx, Cy) in the calculation, assuming that the nose tip is a blocked polygon for each identifiable cross section contour. The calculation is then done for the total C = (Cx, Cy) contour data of the mean value of centroids. This average centroid point is calculated by the vertical axis located at the distance of the average boundary point. The maximum distance corresponds to the tip of the nose. 7 (a) and 7 (b) show a description of an example idea of the present invention. Centroid terminology refers to the geometric center of the shape of an object, ie the center of mass or the center of gravity, as described above.

It can be computed from the point of view of (X, Y) defined by the non-overlapping centroid's closed polygon N vertices and the following vertices:

(Equation 4)

Calculate as above in area A.

Centroid is set to C = (Cx, Cy)

(Eq. 5)

(Equation 6)

The polygon must be closed and the vertices (Xn, Yn) must be equal to (X0, Y0).

<Feature Extraction>

The centroid distance that has been used in feature extraction contour search applications. The centroid distance for contour J is defined by Equation 7 below.

(Eq. 7)

Cj≡ (Cjx, Cjy) is the centroid of the Jth contour, and (Xjx, Yjy) is the boundary point. 9 shows an example of the calculated centroid distance.

Although centroid distances can be used in functional vectors, in addition, the number of boundary points of vector processing may not be equivalent for gallery and probe images. So once the centroid distance vector is evenly calibrated using interval 256 point boundary spline interpolation. The magnitude of this vector also shifts to account for the change in position of the DFT boundary point. Thus, the final function for authentication can be obtained.

[<Performance Analysis]

<Match score>

The cross section produces 21 function vectors every 256 lengths. These are immutable functions. The 256 * 21 matrix Rij. Is created by collecting the function vectors. The definition of the match score 0 ≦ S ≦ 1 is shown in Equation 8. (See FIG. 8).

(Eq. 8)

Symbolizes probe and gallery images inserted below P and G. And '/' stands for matrix area.

<Result>

400 data samples were collected. One of them is of low quality and is not used. Each sample is taken one from the gallery data per subject and the rest is used as a probe. The FAR and FRR shown in FIG. 10 show different thresholds. FAR decreases rapidly at 0.91.

The method according to an embodiment of the present invention uses 100 data sets and 399 full data sets. The high number of studies compared most with previous studies. If most of the past research on this work is about verification, this work has focused on certification. EER is an impressive 4.563% considering the size of the data. The error rate of the authentication is expected to come out low based on the geometric surface. This happens behind data spikes and holes. This error is the main reason for the relatively high EER. There is no effort to minimize the impact from these errors. Other reports in Table 1 include identity verification / authentication methods. This is summarized in "49".

Table 1 below compares the method of the present invention with the conventional method.

As discussed above, face recognition has been the most widely studied and investigated for the last 15 years. A recognition method is proposed based on the image of a group of 2D and 3D. However, large data sets have yet to achieve very high success rates. This chapter reports on three-dimensional surface shape data. The proposed method extracts a pose, which is a function for authentication, by using cross-section contour data during authentication. The graph flow of the proposed method can be seen in FIG. 4.563% of EERs were achieved in the set data of 100 individuals.

[Checking Hand Shape Using Radon Deformation]

Many physiological and behavioral characteristics of humans generally do not change over time and are unique to each individual. These biometric features include faces, irises, fingerprints, hand geometries, and palm prints, which are used for secure access. The most widely studied biometric features are 2D and 3D face recognition and fingerprint recognition. In the case of the face, there are problems due to the reliability of the living body, pose expression and lighting. Fingerprints have been used as the most effective method of biometrics for decades. However, it also has the limitation that users of large groups, elderly and factory workers, etc. do not provide good quality fingerprint images. The surface of the fingerprint is very small, but if the cuts or scar marks on the surface compromise the integrity, problems with the system occur.

As a result, there is a growing interest in other biometric properties. In recent studies, many researchers have developed the human gait and iris patterns. But the simplicity of hand geometry was more favorable. Human hands use length, width, thickness, palm shape, shape and geometrical characteristics of the fingers. Hand image acquisition is a simple task that does not require any special lighting or resolution. Low resolution images can also be used in palm fingerprint authentication systems. On the other hand, there is also an integrated identification verification system that can be used in verification systems with high-resolution images. Some patents allow for personal authentication using the features of hand geometry. In order to achieve effective personal authentication, we have designed a system using the bootstrap technique by utilizing the geometrical features. A similar system is described in <83> using the hand geometry function. Some related work appears in digital hand images <77, 84, 85> using low resolution images. It is used when it is possible to obtain a result of limiting hand movement on a fixed sphere. It can also be biased to collect small databases, and fraudsters can easily break into systems using fake hands.

The method for a web security system is a development verification system based on prototypes of hand figures. We tested an image set consisting of 360 images and obtained 2% FAR and 15% FRR 15%. The hand shape verification system used a database of 200 hand images from 20 people, with a 97% error rate below 10% for identification and verification. Oden et al. Reported a system using implicit polynomials for identification and verification. They combined their method with identification, with 95% success and 99% of verification successful geometric features. The training set they use consisted of 40 images. Jain and Duta developed a verification system based on finger contours and measured alignment errors between them. They experimented with 353 images from 50 people and reported 2% FAR and 3.5% FRR.

<Motivation>

A convenient editing system and a good identification and authentication system come from the popular hand shape of biometrics. Previous work has involved a variety of methods including length, width, thickness, geometry, palm shape, and finger geometry for shape recognition. Pegs, a typical method, are almost always used to correct the position of the hand and are used to shoot with anatomical measurements <77, 84, 86>. In the hand outline and extraction function, it is represented by a group of points that emits a role in the verification process. The purpose of this study is to suggest Zernike moments in high order as a means of confirming the position and position of the hand shape. However, we use the Radon transform to extract shapes by hand from the proposed method in this study. The palm of the hand has a very important anatomical function and this method is used. The distance between the center of the palm's mass and the boundary is the maximum at the tip of the middle finger. This fact is used to find the optimal parameters for the radon transformation and the ability of the two hands to look for functions that are invariant in one dimensional position.

<An example method according to the present invention>

The non-peg handed authentication scheme was proposed in this study, which uses the Radon transform to decompress the invariant functional vector position. The radon transformation simply represents the model image of the object as a line from 0 to 360 degrees. However, depending on the direction of the radon transition, optimal shooting is possible. The optimal angular position should find the centroid of the palm and the tip of the middle finger. This section also describes the various steps of the feature extraction process. The discussion begins with a brief introduction of the radon transformation.

<Radon transformation>

Named after Austrian mathematician Karl Johan Radon August, famous for his radon transformation.

One major application of radon is CAT scan conversion. It is also used to frequently search for lines of an image. The radon transformation of one image f (x, y) is a one-dimensional function in which the given angle is zero and transforms by projecting the image along the given angle and the resulting prediction sums the key lines of pixel concentration in that direction, or in another word. .

The radon transformation of the F (x, y) image can be defined at Eq.

(Eq. 9)

Here, (x, y) symbolizes space and p symbolizes the radial distance from the coordinates of θ angle. The relationship between the spatial coordinate system and the parameters is shown in FIG.

This property is used to detect lines in the image <89>. Line sensing in FIGS. 12 (b) and (c) is shown.

Radon transformation has several important properties. The most important attributes are linearity (Eq. 10), mobility (Eq. 11), and rotationality (Eq. 12).

(Eq. 10)

(Eq. 11)

(Eq. 12)

<Image Collection>

The feature extraction process begins with collecting hand images. The document scanner was used to collect low resolution images on a dark background. 75 dpi images were taken in 8-bit grayscale format. Each image is 638-876 pixels in size. 13 shows an image of an object. A total of 136 images were taken of 18 subjects using a single subject. The gray scale image is then converted by the binary image threshold. The wrist is erased using morphological operators and then the end of the hand's boundary is traced to calculate the centroid. Theoretically, the maximum distance at the boundary of the centroid is the middle finger tip. Together with centroid, this view is used to determine the optimal angle for radon transformation.

<Grayscale binary conversion>

FIG. 14B is a bar graph of the intensity of the sample image of FIG. 14A. The image in (a) was converted to dual format before it was in the range (0,1) before it was calculated as a bar graph. Setting the threshold is easy because the image was shot against a dark background. As you can see from the image, the gray pixels are well separated from the background image. Century span is 0.15. Thus, the threshold was determined to be T = 0.15, which applies to all images in the data. If any intensity value is greater than T, the value is displayed as 1 and if it is lower than it, it is set to 0. Figure 14 (c) shows a binary image. You will see that the dots are white with a dark background. These dots will go away while you remove your wrist.

<Clear wrist process>

Knowing the exact location of the centroid of the palm of the hand can be a success. You must remove the wrist from the image of your hand. This can be done in morphological operations. In morphology, binary image is defined as in Equation 13.

(Eq. 13)

Erase gray pixels smaller than the component in the morphological opening. The component of the disc type is selected within radius 100.

Fingertip check

Checking the middle finger tip is an important process in this method. Together with Centroid, it is the center of the palm and is used to measure the angular parameters of the radon transformation. That is, it provides key information about which direction the line should be shot at these two points. The centroid distance of the palm is calculated by equations 5 and 6. The maximum centroid distance is the hand boundary and the middle finger tip. This point is called the control point in later sections.

This point is symbolized by (CTRLx, CTRLy). To calculate the centroid distance, you first need to track your hand boundary. Boundaries are tracked using eight connected neighbors. This process is described in detail in <90>. Fig. 16A shows a binary hand image. In FIG. 16B, the centroid and the control point are marked with red dots. The control point is located at the maximum centroid distance as shown in FIG.

<Feature Vector Extraction>

The optimal angle for the radon transform can be found by simply calculating the slope connecting the lines (Eq. 14) (Cx, Cy) & (CTRLx, CTRLy), which can be calculated mathematically.

(Eq. 14)

Radon transform calculations of binary hand images will give invariant function vector rotation. However, position invariance cannot be achieved due to changes in other probe images. This problem is written using (Cx, Cy) ≡ (X0, Y0) with θoptimal using the shift property of Equation 6. In FIG. 17, two different hand images have been calculated.

[Performance Analysis]

<Match score>

The match score for the function vector is defined by Equation 15 below.

(Eq. 15)

Here, X and Y represent each gallery and a probe function vector.

<Result>

A total of 136 images were captured. There are a total of 118 probe images leaving one gallery image per person. Another threshold distribution of FAR and FRR is shown in Figure 18 (a). The intersection is threshold 0.82. In FIG. 18B, the ROD curve is shown. EER has 5.1% when the threshold is 0.82.

 In <77> 360 images were tested and 2% FAR and 15% FRR were measured. In <84>, 200 hand images of 20 people were tested and reported that 97% of successful identification and error rates were less than 10%. Both of these experiments are excellent results. Oden et al. (85) showed a 95% success rate and a 99% success rate during the verification process using 40 images. However, the success rate is high but the number of people is not large. Jain and Duta have developed an authentication system that takes 353 images and measures finger contours and misalignments between them. They reported a 2% FAR and a 3.5% FRR. In the study shown here, the EER was 5.1%. FRR was less than 10% when compared to the study in <77>. Also, the verification result was better than that of <84>. However, <86> showed high performance. Although the 136 image dataset used in this study is mediocre, we expect similar results to be seen in larger datasets.

Summarizing the example described above, a new technology for handheld authentication systems is introduced in this chapter. Instead of measuring width and length, this method takes into account several factors that are invariant in hand shape. It can also be run without peg (sphere). A simple document scanner on low-resolution images showed a recognition error rate of only 5.1%. The perspective of the proposed method is shown in the figure. 134 images were tested. However, performance analysis requires larger data sets.

[Biometric System using Force Profile]

Gait is one of the characteristics required for newly recognized biometrics. Difficult to imitate or difficult to follow is one of its advantages, and this is the only one that can identify people even when they are far away. The study of gait recognition can be briefly divided into two main classes. The first is the state space method, which is a pose in which the gait is configured in the order of the body, and is a recognition method obtained by observing a static pose <92,93>. Murase and Sakai <92> studied a template matching method that distinguishes different steps in their own space. Huang et al. (93) have added authoritative analysis. The second class is space-time method. The space-time method is in the continuity of operations <94, 95, 96>. In <94>, he applied a curve fitting “snake” based on the space-time gait pattern when others walked. Little and Boyd <95> took advantage of changes in gait according to the frequency and pattern of optical flow. Recently, Shutler et al. [96] measure gait perception widely from state-space to space-time. BenAbdelkader et al. (91) are studying the steps of measuring and measuring the self-similar image of a moving person.

Existing gait recognition algorithms are designed to verify identity over distance. Most gait silhouettes have been extracted from moving images. These images are captured in a controlled environment. There are many proposals for viewing angles. Recently, however, a new approach to gait recognition has been studied. Rather than studying the shape of the walking in the image sense, the new method studies force profiles or walking acceleration. Force profiles are thought to vary from person to person. Accelerometers are used to capture accelerometer data for authentication <97>. In [98], several gait and complex classifications extracted from force profiles have been used for recognition.

It is suggested to use the gait profile from the force measurement board as a way to verify your identity. The advantage of this system is that it can be authenticated by human gait if no other biometrics are possible. Another advantage is that this recognition method is a modest recognition method than other biometric methods. Every individual has a different step. In the proposed method, a lot of data were collected using the gait data from the force measurement board. Depending on the force measurement, three different ground reactions are available: vertical, horizontal and lateral. With the help of DWT, data can be collected uncontaminated. Then the similarity of the histogram for the recognition goal is used.

[Data Extraction and Preprocessing]

<Data extraction>

Ground reaction force data was extracted using an AMTI force plate. Many of AMTI's platforms are built on biometric laboratories around the world and hundreds of steps. In three-dimensional space, force and time can be divided into three components around three axes. However, the force measurement platform provides six outputs that react with three orthogonal forces, and three orthogonal moments react in the platform's appearance. The three tables in the coordinate system are called X, Y, and Z, and the associated forces and times are called Fx, Fy, Fz, and Mx, My, Mz. This force platform is a positive Z-axis pointing down in the right coordinate system. The positive Y axis is horizontally forward and the positive X axis is centered to the left and looking towards positive Y (99). This coordinate system is shown in FIG. In this figure, cross means that the Z axis is vertically downwards.

Three men and three women aged 22 to 45 years of age participated in the study. The purpose of this study is to explain the purpose of the study to each subject before entering the experiment. Participants practiced enough to walk steadily under normal conditions. Each participant repeated the same experiments, resulting in three clean experiments. The first data sample was used as registration data and the other two were used during the recognition process. Therefore, we participated in two substantive experiments and 10 fake experiments for each participant and received a total of 12 substantive results and 60 fake results. This data was collected from the Motion Analysis Laboratory of the Korean Institute of Standards and Science's Health Measurement Group.

<Data preprocessing using ripples>

In the force measurement plate, the signals of the X, Y, and Z axes can be seen in FIG. 21. It is possible to collect and analyze all data from three directions or to combine multiple signals. However, our study found that Z-axis data is the best data for recognition. This raw data can be contaminated by the noise of the surrounding environment or the sound from electronics or digital computers. Noise becomes a big problem once we reach the practical recognition process. So we decided to protect the data through the ripple transform. The individual ripple transform signals of X are defined as follows (Eqs. 16, 17, and 18).

(Eq. 16)

The wavelet coefficient is defined as follows.

(Eq. 17)

(Eq. 18)

Level 5 wavelets use ‘db4’ to prevent noise. So we know that if the noise is zero, it means white Gaussian noise. Threshold decomposition at each level is chosen with a smooth threshold. The selected threshold value follows the principle of Stein's Unbiased Risk estimate (RISK). If we model the noise signal as the sum of the clean signal and the white noise, the details of each level will overlap the clean signal and noise details. Applying white Gaussian noise and a soft threshold will reduce the noise noticeably. 22 and 23 show gait data of general Z-axis and data after noise reduction.

<Recognition Algorithm>

The bar graph is Hi-mapped and counts the measured data samples. The data may fall into categories that do not have anything in common. Therefore, if N is the total number of samples, then K is the total number of bins and histogram Hi is given by Eq.

(Eq. 19)

Normal data histograms normalize the number of samples going to the trash out of the total samples. In other words, the normalized bar graph is in the distribution of waste bins to the bar graph. Mathematically, if ni is a population of ith, its normalized bar graph is given by Equation 20 below.

(Eq. 20)

To extract the function, normalize the n reservoir histogram to the Z-axis plate, as shown in Figures 24 and 25. The advantage of the histogram is the amplitude of the data signal, so phase delay is not a major concern in the recognition process, as shown in Figure 24 in this experiment. Normalization of the N-trash histogram means that data samples in all ranges are divided into equal sizes, and the value shown on the vertical axis in FIG. 25 is the percentage of each bin in the total number of samples. If the histogram shows very similar results for all three experiments, this is a good method for the recognition system. Therefore, we need to define two histograms for the price index in terms of absolute distance.

(Eq. 21)

Xi is a case where data is dropped into a trash bin during the normal registration of Histogram X, and Yi is a case where data is dropped into a trash bin during the verification process. N is the number of bins used to compute the histogram. The distance value (d) represents a similar score between two fake data samples. The more similar the two histograms, the smaller the value of d. It is normal that the actual user's attempt value is lower than the fraudster's attempt value. Figure 26 is a brief description of the processing procedure of the presented method.

<Result>

FAR and FRR are used to analyze the performance of the proposed method, and the formulas defined are Equations 2 and 3.

Lower FAR and FRR mean better perception. Table 1 shows that FAR and FRR have values for different bin numbers. N = 5, 10, 15, 20, 25. N = 5, FAR is 3.33% and FRR is 16. This gives the best results from the histogram method. Another important observation is that recognition rates are not very high when N = 15 or above (see Table 2).

Biometric force-based force plates can be run in sensor-based authentication mode like a gait authentication system. Unlike vision-based gait recognition systems, surveillance cameras and images can be used to identify people and, in the case of their gait or force plate biometrics, users can be identified with a limited access system. Force measurement data is a viable option if fingerprint or facial recognition systems are not available. The system can also be used with fingerprint or facial recognition systems. The force measurement biometric system introduced here is much simpler than previously introduced. At this time, we used one function, but in the previous study, three different functions were used in the verification process. In addition, the existing ones required complex classification for identification. However, we use a simple histogram based on distance measurement for the verification process. Although their analysis was conducted on 11 doubled subjects, we expect similar results to be achieved in our more successful and much larger experimental groups. In this paper we analyzed only one step signal. A similar analysis of multiple steps data will yield a more successful recognition rate.

[summary]

A big issue today in biometric systems through gait is the one-step authentication method. This method extracts from the force plate through ground reaction forces. The advantage of this recognition technology is that the user can use it easily and conveniently. In the present invention, we present several evidences of the authentication method based on one step. The research results can be divided into two. The first is noise reduction test data, using wavelets to reduce db4 to level 5. The histogram then confirmed the individual's steps. The results are encouraging for a small group. In the future, however, experiments need to take into account variables such as problems with larger groups, such as wearing different shoes or carrying luggage. It is also necessary to run the algorithm to reduce the error rate.

 [Conclusion and Direction to the Future]

The present invention relates to biometric authentication and verification. Biometric Authentication. Verification Users have been the most studied over the last few decades. Today, advances in computer, graphics, network, and database systems and digital signal processing have increased the market for biometrics. Emerging biometric research includes gait, ear, and instep, and these characteristics are being studied alongside existing face, fingerprint, and hand shapes.

In the present invention, three new bioschemes have been dealt with in three different fields. A general introduction to biometric systems was first introduced. Each of these three methods was subsequently introduced. Among the biometric features introduced in this paper are face, hand and force profiles. Algorithms are presented and performance is evaluated.

In Chapter 2, I wrote about new face recognition technology using 3D surface theory. This method divides the contours into a horizontal model with face surface data. The area between the nose and the eye is then used to create a one-dimensional shape. This descriptor is used during the authentication process. The algorithm tested is 100 people and 299 "probe" images. In this method, EER (when FAR and FRR are the same) showed 4.563%, which is very bright for this size dataset. But in surface science, noise is a factor that increases error rates.

The next chapter introduces a new authentication technique for hands. Low resolution hand image captured by a document scanner. It does not have a Peg and captures binary hands using the Radon transform as a method of extracting invariant poses. This method was tested using 136 "probe" images of 18 people. This recognition method has an EER of 5.1%.

Phosphorus profile is also introduced in Chapter 4. The force profile of the subject is that, when a person walks on a pressure sensitive mat, the force is measured on three different axes. If the X-axis is used in the force data, noise reduction is performed through wavelet transformation. The histogram from which the noise-free data is calculated is then the distance index. Different numbers of bins are used in the recognition process. The best results are when FAR = 3.33% and FRR = 16% with five bins.

The biometric scams presented in this study have good results. However, there are certain aspects of couples and future research that should be further improved. Noise is the biggest problem in the face recognition process. In this study, using 3D surface data, no attempt was made on noise. This is a problem that must be raised to improve performance. In addition, a combination of vertical and horizontal contours will provide better performance.

In the hand shape verification method, the finger length, width, palm width, size, etc. can be combined with hand geometry to improve performance. In another potential study, a mixture of palm prints and hand shapes can be used to study multifunctional recognition methods.

The force profile can be studied with time. You can also capture and time your shoe heel and toe drop and include it in the recognition process. Human stride length can also be included in the force profile. Another possibility is to capture and create three force profiles: flat, uphill and downhill. However, these experiments are very difficult to design.

Three new biometric patterns are discussed in this paper. Promising results are included and will certainly contribute to future research.

1 illustrates the main components of a typical biometric system.

2 illustrates an example of a ROC graph model.

3 illustrates an example of several biometric features.

4 shows four different postures according to an apparent database.

FIG. 5 shows a landmark marker (a) in a three-dimensional face and a three-dimensional shape (b) from which markers and information have been removed.

6 is a cross section contour extracted from a three-dimensional shape from which markers and information are removed.

7 shows a demonstrating nose tip procedure.

8 shows four selected contour function extractions.

9 is a graph illustrating data for setting different threshold values between an actual user and a fraudster.

10 is a graph showing the work characteristic curve (a) and the decision flow curve (b) of the prisoner.

11 is a diagram flow diagram of the present invention.

12 illustrates a change in radon according to an embodiment of the present invention.

Figure 13 shows the collected hand image according to an embodiment of the present invention.

14 illustrates grayscale binary conversion according to an embodiment of the present invention.

Figure 15 shows the state after successfully clearing the wrist portion in accordance with an embodiment of the present invention.

16 illustrates an example for calculating the centroid distance after erasing the wrist part according to an embodiment of the present invention.

17 illustrates an example of a compressed position of an invariant function according to an embodiment of the present invention.

18 illustrates another grain boundary value distribution (a) and an ROD curve (b) of FAR and FRR according to an embodiment of the present invention.

19 illustrates a flow diagram according to an embodiment of the present invention.

20 illustrates X, Y, and Z axes associated with a foot in accordance with one embodiment of the present invention.

FIG. 21 is a graph illustrating general force measuring plate data using three coordinate axes according to an embodiment of the present invention.

FIG. 22 illustrates gait data along the Z-axis before the wavelet removes noise according to an embodiment of the present invention.

FIG. 23 is a diagram illustrating gait data along the Z axis after the wavelet is noise canceled according to an embodiment of the present invention. FIG.

24 is a graph illustrating three times the force measurement plate data of the X axis of the object 1 according to an embodiment of the present invention.

FIG. 25 illustrates a ten store histogram corresponding to FIG. 24.

Figure 26 illustrates a procedural step of finding histogram similarities according to an embodiment of the present invention.

Claims (4)

Subdividing the head data using the marker position; Extracting a contour by separating cross sections from points on the surface of the collected head data; Identifying a nose tip identification of a region of interest from a contour map obtained from the extracted contour, and extracting a cross-section contour; After extracting the contour, correcting the centroid distance vector using boundary spline interpolation using a function extraction contour search application; And And correcting the centroid distance vector, and performing a performance analysis to define a match score. 3. The method of claim 1, Extraction of the cross-section contour is a three-dimensional shape measurement system for producing a three-dimensional object, characterized in that the following formula 3 to 5. (Equation 4)

Calculate as above in area A. Centroid is set to C = (Cx, Cy) (Eq. 5)

(Equation 6)

The method of claim 1, Centroid distance vector using the feature extraction contour search application is a three-dimensional shape measurement system for producing a three-dimensional object, characterized in that defined in accordance with Equation 7. (Eq. 7)

(Where C _j = (C _jx , C _jy ) is the centroid of the J _th contour, and (X _ji, Y _ji ) is the point of the boundary.) The method of claim 1, The collection of the head data is a three-dimensional shape measurement system for producing a three-dimensional object, characterized in that for using the Radon transform to represent the model image of the object by a line from 0 to 360 degrees.

Images