The present invention relates to a technology for identifying a person using a living body, and more particularly to a technology for identifying a person using a finger vein pattern.
Today, the typical personal identification technology is fingerprint identification. However, the problem is that other person's fingerprint is easily obtained, for example, a criminal's fingerprint is taken in the scene of a crime, and therefore a finger-print may be forged. This problem leads to the development of personal identification technologies other than fingerprint identification. For example, JP-A-7-21373, laid-open Jan. 24, 1995, discloses a personal identification technology thorough the use of a finger blood vessel pattern, and JP-A-10-295674, laid-open Nov. 10, 1998, discloses a personal identification technology through the use of a vein pattern on the back of a hand. These technologies shine a light on a finger or on the back of a hand, capture the reflected light or transmitted light, extract the blood vessel pattern from the captured image, and compare the captured blood vessel pattern with the previously-registered blood vessel pattern to identify a person.
However, there are some problems in implementing a personal identification system that uses finger vein patterns.
One of the problems is the reproducibility of a captured image. Although a conventional personal identification system has positioning parts such as a pin or a grasping bar for stabilizing the imaging region, an error in the imaging region is unavoidable, for example, when a finger is rotated or moved in the plane or when a finger is rotated on its major axis. Therefore, it is difficult to completely match a registered vein pattern with a vein pattern obtained at identification time, with the result that the performance of identification is reduced. In particular, on a fully-non-contact system on which the finger is not put on something for fixing, a registered vein pattern and a captured vein pattern may differ largely and this difference further reduces the performance of identification.
Another problem is a light source. A conventional personal identification system has no function to adjust the amount of light from the light source. This generates several image-quality problems such as a blurred outline of a captured image, a lack in sharpness, and a low contrast. These problems require a complex image-processing algorithm for correction and sometimes result in the low performance of identification.
According to one aspect of the present invention, means described below is used for improving reproducibility. First means is an algorithm for correcting an error detected during image processing that is executed for matching an imaged finger blood vessel pattern with a registered pattern. This correction prevents the performance of identification from being degraded.
Second means is a three-dimensional imaging of a living body from various angles using a plurality of imaging devices. Even if a registered vein pattern was imaged from only one direction, that is, the registered pattern is two-dimensional data, the finger may be placed freely when imaged for identification. Therefore, even if there is an error in the imaging region, one of the plurality of images may be selected for use in matching. This prevents the performance of identification from being degraded. It is also possible to register three-dimensional vein patterns as patterns to be registered by imaging the vein pattern from a plurality of directions. In this case, one of the plurality of registered vein patterns is selected for matching. This also prevents the performance of identification from being degraded even if there is an error in the imaging region.
Combining the first means with the second means further increases the performance of identification.
A light source with means for optimizing the amount of light at imaging time is used as the light source. This configuration optimizes the amount of light from the light source to make the quality of a captured image best.
As a vein pattern for use in personal identification, it is more preferable to use the vein pattern of a palm-side finger than to use the vein pattern of a back-side finger. This is because the vein pattern of the back-side finger, which is always exposed externally, is more likely to be stolen. In this embodiment, the system always images the vein pattern of the palm-side finger.
A plurality of light-emitting devices, which make up the light source, are arranged according to the shape of a finger, as shown in
The procedure for optimizing the amount of light is as follows. When human being's finger or toe is imaged, the light transmission factor is highest in joints. Therefore, the system detects a joint from the light intensity profile of a finger in the major axis in the image data and uses the maximum intensity value as the intensity value (B) of the joint. This value is compared with the reference value (A) of intensity that is previously set. If A−B<0, the light source is subjected to a feedback to reduce the input current to the light source. If A−B>0, the current input to the light source is increased. When A−B=0, the system ends the adjustment of light intensity, captures the image, and starts image processing. Performing this processing for each light-emitting device optimizes not only the amount of light but also the area of the light source. In this case, the light source with the configuration composed of chip-type, small LEDs arranged in the plane, such as the one indicated by numeral 405 in
The method described above adjusts the light source output to optimize the light intensity. Alternatively, adjusting the time during which the light stays on may also optimize the light intensity.
To optimize the amount of light, it is required that the joint be identified. Two sample procedures for identifying the joint will be described. In one procedure, a portion with a relatively high light-intensity value is detected in the image profile of a blood vessel pattern of a finger, imaged through the use of transmitted light, to identify the joint of the finger. Then, a feedback is effected such that no pixel reaches the intensity value of 255 in the 8-bit dynamic range. In another procedure, an image to which a spatial low-pass filter is applied in the major axis direction of a finger in the captured image is evaluated, and the amount of light of the light source striking the joint is adjusted. Any of the procedures described above forms a light source with a spatial intensity distribution.
The above-described configuration for automatically adjusting the amount of light is suitable for capturing a high-contrast blood vessel image. Adjusting the amount of light significantly increases the quality of a blood vessel image, allowing person identification through image-to-image operation of captured images to be performed smoothly.
With a previously registered finger blood vessel as the template, personal identification operation is performed through the calculation of correlation to find a similarity between the blood vessel pattern image of a finger imaged at authentication time and the template. The calculation of correlation is a monotone increasing calculation in which an output value increases in proportion to the degree of matching of two-dimensional array elements. Most typically, a two-dimensional convolution calculation (formula 1) is used.
1z(k1, k2)=mn×(i,j)y(k1+1−i, k2+1−j)(k1=1,2, m+n−1, k2=1, 2, m+n−1) (Formula 1)
One to ten fingers, usually up to all fingers and toes may be registered. Depending upon the required security level, the number of fingers to be compared may be increased. In some cases, a non-finger blood vessel image may also be used with a finger image.
During registration processing, the image detecting means captures a person's finger image to be registered (block 1000). At the same time, registration image creation processing, which will be described later, is performed to create a finger-vein emphasized image and the created image is registered (block 1001). On the other hand, during authentication processing, the personal information receiving means receives a person's ID (block 1002) and, at the same time, the registered image corresponding to the received ID is selected from the database (block 1003). In addition, the image detecting means captures an identfee's image to be authenticated (block 1004) and, at the same time, authentication image creation processing, which is similar to registration image creation processing and will be described later, is performed to create a blood vein emphasized image (block 1005), and the calculation of correlation between the captured image and the registered image is executed.
Then, the result of the calculation of correlation is evaluated, and the authentication result indicating whether or not he/she is an identical person is output. Most typically, a two-dimensional convolution calculation is used as the calculation of correlation. In this case, even if the finger is translated in the image plane, the distribution obtained as a result of the two-dimensional convolution operation is also translated with no change in size and shape. Therefore, the evaluation of similarity between these two images automatically corrects errors generated by the translation in the image plane. In addition, taking advantage of the fact that the convolution operation between two data units is equivalent to the inverse Fourier transformation of the product of the Fourier-transformed data units, two-dimensional Fast Fourier Transform (hereinafter called FFT) may be used to speed up the calculation of correlation.
The image to be registered and the image to be authenticated, for which finger-edge extraction and image-rotation processing has been performed, are each converted to a finger-vein emphasized image (block 1102), and the two-dimensional FFT operation is executed for the converted result (block 1103). The result generated for the former image is registered. The result generated for the latter image is multiplied by the registered image selected based on the received ID (block 1104). Then, two-dimensional fast inverse Fourier transformation (Inverse FFT, hereinafter called IFFT) is performed for the result (block 1105) to produce the correlation distribution of the registered image and the image to be authenticated. As described above, the same image processing is performed for blocks 1101 to 1103 during registration image creation processing (block 1001) and during authentication image creation processing (block 1005).
The personal identification system may include a light source which shines a light on the imaging region of a living body, an imaging unit which detects a transmitted light from the imaging region to image the living body, and an imaging processing unit which extracts the blood vessel pattern of the living body from the image converted by the imaging unit and compares the pattern with a previously-registered blood vessel pattern, wherein the image processing unit may comprises means for correcting an error between the registered blood vessel pattern and the imaged blood vessel pattern.
An input image is divided roughly into a finger 120, an edge 121, and a surrounding part 122. In general, the image also includes various noises 123 that must be removed. In
In response to a captured finger image (block 1200), a high-cut filter filters out small components such as noises 123 (block 1201) and emphasizes only relatively large components such as the edge. Then, the procedure executes directional differentiation (block 1202) to give an edge-enhanced image (
As shown in the image in
In the image shown in
When the average of the pixel values inside the finger is shifted to 0, the pixel value of 0 is inserted into the surrounding part 122 of the image shown in
Next, after creating a difference image between the original image (
Finally, according to the inclination of the finger obtained from the detected edge location, the image is rotated such that the finger is inclined at a fixed angle, typically, at an angle of 0 degree (
Cab(x, y)/({square root} {square root over (Caa(x,y))}.times.{square root} {square root over (Cbb(x,y))}) (Formula 2)
where, Cab(x,y) is the correlation distribution of the registered image and the image to be authenticated. Caa(x,y) and Cbb(x,y) represent a sum of squares of respective pixel data of a registered image and a sum of squares of respective pixel data of an image to be authenticated, respectively.
If the calculated maximum correlation value M of the registered image and the image to be authenticated is larger than the threshold Mo, the identifee is regarded as valid and is accepted (block 1009). If the value M is less than the threshold Mo, the identifee is not regarded as valid and is rejected (block 1010). The threshold Mo should be statistically calculated in advance by entering sample images. When the average of pixel values inside the finger is shifted to the value of 0, the value of Mo ranges from 0.45 to 0.55. However, the value is not limited to this range.
If a person is not acknowledged, he or she must re-enter data, such as finger image capture data, to make a re-authentication request a predetermined number of times. For example, a person who is successfully acknowledged is allowed to access managed data or areas. On the other hand, a person who is not acknowledged makes a re-authentication request a predetermined number of times and, if the person is not yet acknowledged, access to the managed data or areas is rejected.
It is desirable that the personal information input means, which is used to select a person's registered finger image from the database, not be a keyboard but a non-contact device. For example, personal information such as a name or a password, or a keyword that is known only to the person, may be input via voice or, alternatively, stored on a non-contact IC card. This type of input means makes it possible to build a personal identification system that takes advantage of non-contact features. This processing may be done independently by the CPU of the identification system or by an online system via computers.
An image to be authenticated is stored on a fixed medium connected to the authentication server, a medium containing semiconductor memory, or a portable medium such as a floppy disk. This method eliminates the need for keyboard operation on a banking ATM, solves the problems associated with a contact input device, and relieves a maintenance nuisance. The advantages described above make this method suitable for gaining access to personal information in an electronic government or for authentication in online transactions.
In the first embodiment, a finger is imaged with one CCD camera. A finger may also be imaged with a plurality of CCD cameras during authentication or image capturing to increase the performance of identification.
Not only a still image but also a moving image may be imaged. When imaging and registering a three-dimensional moving image, a finger is rotated in the configuration, shown in
The image is captured into the processing unit for use in image operation and authentication. Because the most similar vein pattern to be registered or the most similar vein pattern to be authenticated is selected for authentication, this method is advantageous in that the performance of identification is increased and the image operation load is reduced.
For example, blood pattern enhancement processing is performed using a filter, such as the one shown in
Two-dimensional FFT transform operation is performed for the obtained image (block 1103), squaring of respective pixel data of the obtained is performed (block 1304), and then two-dimensional IFFT operation is performed (block 1305). For the obtained result, the parameters for evaluation are calculated (block 1306). Because vessels usually run in the major axis direction of the finger rather than in the minor axis direction, the difference in the blood vessel pattern is most reflected in the peak shape in the minor axis direction of the finger in the two-dimensional convolution operation result. Therefore, the parameters for evaluation are calculated, for example, using a kernel composed only of the elements in the minor-axis direction of the finger such as the one shown in
If the input image is an m.times.n matrix and the kernel is a p.times.1 matrix, then the result of the calculation of parameters for evaluation is a (m+p−1).times.n matrix. The maximum value Mx of the resulting matrix is calculated for each of the registered image and the image to be authenticated. Let the maximum value be M1 and M2, respectively. M1 is stored in the database (100).
M=M.sub.12/{square root} {square root over (M.sub.1.times.M.sub.2)} (Formula 3)
If the calculated value MX of correlation between the registered image and the image to be authenticated is larger than the threshold Mxo, the identifee is regarded as valid and is accepted (block 1009). If the value Mx is less than the threshold Mxo, the identifee is not regarded as valid and is rejected (block 1010). The threshold Mxo should be statistically calculated in advance by entering sample images. When the average of pixel values inside the finger is shifted to 0 as described above, the value of Mxo ranges from 0.3 to 0.4 but is not limited to this range.
Because a fully non-contact method is not always advantageous in cost, processing time, and compactness, it is more practical for a device, while still retaining the non-contact features described above, to have the minimum positioning parts required for fixing an imaging region such as a finger or a hand. Note that more bacilli are present on the palm of a hand of a human being than on the back. Therefore, even on a device on which the imaging region contacts the positioning parts, the palm of the hand should not contact the device. The following describes an example.
An optical sensor, which measures the distance between the device and the palm to check to see if the height, from the device to the palm, is correct, is provided to control image capturing. If the height is incorrect, incorrect-height information is sent to the identifee. Because the palm does not contact any object as described above, this embodiment reduces the possibility of bacillus contagion caused by an unspecified number of persons using the device. Therefore, it can be said that the method in this embodiment in which the palm does not contact any object is better than a method in which the palm contacts the device.
Winding a palm-contact sheet (900) or sterilizing the sheet with a ultraviolet light source (901) or a disinfectant (902) allows even a palm-contact device to take advantage of the identification system according to the present invention that keeps the device clean. For full sterilization, an optical-catalyzed (titanic-oxide) coated sheet is used as the palm-contact sheet (900) and a ultraviolet light is shown on the sheet. Including this type of sterilizer keeps the device clean.
The embodiments described above allow a reliable, secure, and easy-to-use personal identification system to be built. That is, a familiar, forgery-proof, highly-accurate personal identification system may be implemented while eliminating or minimizing maintenance management work executed for preventing contagion caused by dirt on the device or for preventing errors in obtained data.
Number | Date | Country | Kind |
---|---|---|---|
JP-2000-290325 | Sep 2000 | JP | national |
This application is a Continuation of U.S. patent application Ser. No. 12/654,889 filed Jan. 7, 2010, which is a Continuation of U.S. patent application Ser. No. 11/976,681 filed Oct. 26, 2007, which is a continuation of U.S. patent application Ser. No. 11/892,506 filed Aug. 23, 2007, which is a continuation of U.S. patent application Ser. No. 11/104,564 filed Apr. 13, 2005, which is a continuation of U.S. patent application Ser. No. 10/733,396 filed Dec. 12, 2003, which is a continuation of Ser. No. 09/954,067 filed Sep. 18, 2001. Priority is claimed based on U.S. patent application Ser. No. 12/654,889 filed Jan. 7, 2010, which claims the priority of U.S. patent application Ser. No. 11/976,681 filed Oct. 26, 2007, which claims the priority of U.S. patent application Ser. No. 11/892,506 filed Aug. 23, 2007, which claims the priority of U.S. patent application Ser. No. 11/104,564 filed Apr. 13, 2005, which claims the priority of U.S. patent application Ser. No. 10/733,396 filed Dec. 12, 2003, which claims the priority of U.S. patent application Ser. No. 09/954,067 filed Sep. 18, 2001, which claims the priority of Japanese application 2000-290325 filed on Sep. 20, 2000.
Number | Date | Country | |
---|---|---|---|
Parent | 12654889 | Jan 2010 | US |
Child | 13416307 | US | |
Parent | 11976681 | Oct 2007 | US |
Child | 12654889 | US | |
Parent | 11892506 | Aug 2007 | US |
Child | 11976681 | US | |
Parent | 11104564 | Apr 2005 | US |
Child | 11892506 | US | |
Parent | 10733396 | Dec 2003 | US |
Child | 11104564 | US | |
Parent | 09954067 | Sep 2001 | US |
Child | 10733396 | US |