JPH11283033A - Method for utilizing feature amount for image identification and recording medium for storing program therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To identify a registered person himself/herself by using a face image printed on a card in a security system of entering/leaving person management or payment using a credit card, etc. SOLUTION: A face image 3 printed on a card 10 with recorded individual information is read, a normal vector is extracted as a feature amount for indicating the face shape, a deviation amount from the normal vector extracted from the face image and registered in a data base system 11 beforehand is obtained, and an individual identification processing is performed corresponding to the amount. An IC memory 4 incorporated in the card for storing the information of the feature amount/an identification function, a PC 1 for an image processing and data retrieval and a data base 2 inside the system are provided.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for determining whether an image is the same or different by replacing image information obtained by photographing the object with less information while retaining the characteristics of the object. More particularly, the present invention relates to a method of using a feature amount for image identification which makes it possible to distinguish a face image printed on an IC card into a read feature amount, and a face image of a card owner previously read and extracted and registered. The present invention relates to a method for identifying the same facial image as compared with the feature amount of the above, and a recording medium storing the method converted into a program.

[0002]

2. Description of the Related Art Techniques for identifying image information include a method in which the position, size, shape, etc. of components included in an image are extracted as parameters and compared as feature amounts, and a method in which the image density distribution is compared as feature amounts. And how to do it. When divided into a plurality of components for comparison, it is easy to understand and handle, but extraction of parameters requires extremely complicated processing and low robustness. On the other hand, in the latter method, discrimination performance is affected by preprocessing such as alignment, and a large amount of calculation is required. For example, if a perfect match is to be obtained, the identification can be performed based on the position and density information for each pixel on the image, but this is obviously impractical.
In addition, determination is made based on the number of pixels (= area) having the same density as a result of performing density slicing using a threshold. Although very strict determinations can be made by these methods, when the amount of information of an image increases, the amount of calculation for the information becomes very large. Therefore, in order to reduce the amount of calculation, a method of dividing a part of the face image into a mosaic pattern and performing a comparison process using the average density value as a representative value for each block has been performed. Is effective but often loses important features due to mechanical division. Examples of the prior art include “Individual identification experiment for gate management using partial images near both eyes of the face” by Hideyuki Shimada, Hirofumi Isobe, Mitsuru Shiono, (IEICE D-2, Vol.1, J77-D-
11, NO. 9, pp. 1680-1690, September 1994.

[0003]

In order to determine the shape of an object based on image information such as a photographic image, the most reliable method is to verify all information possessed by the image. However, when high-resolution images are generally used, the amount of calculation and the time required for the determination become extremely enormous. For this purpose, there has been reported a method of dividing an image into a mosaic pattern and compressing the amount of information by using an average density value of each mosaic plane as a representative value. However, if information necessary for making a sufficient determination is stored, there is a problem that a large compression ratio cannot be obtained. In addition, the mechanism of the determination is an alternative determination of whether they are the same or different, and giving an intermediate result generally abandons the determination, and a process called rejection is performed. I will be rejected from the judgment system. Therefore, an object of the present invention is to solve the above-described conventional problems and store a method of using a feature amount for image identification that enables stochastic identification determination according to the amount of calculation, and a program thereof. It is to provide a recording medium.

[0004]

In order to achieve the above object, a method of using a feature amount for image identification according to the present invention provides a feature for reducing a calculation amount and a calculation time in an identification process using image information. The amount is determined not by the average of the density values but by the angle formed by a normal vector expressing the gradient of the density value, and is realized by using a processing algorithm adaptive to image information. In other words, if the amount of information that can be a sufficient judgment material and the capability of the computer can be given, it is possible to make the same or different judgment with high accuracy, and if there is only a small amount of information and poor computational power, the probability It is possible to make an unusual determination, and to ask for a final determination by a human, the same or different determination can be made. The portable storage medium for storing personal information according to the present invention is, for example, an IC card or a memory card in which personal information can be written and read, and digital information in which the personal information is coded is stored on the card. A face image for identifying an individual is printed on the card surface. An authentication system for authenticating personal information according to the present invention includes a database system on a network or a local where a large number of personal information is stored in a database, and a card reader capable of reading information from an IC card or a memory card.
An image scanner that reads image information on the surface of an IC card or a memory card, and holds image data;
A storage medium (for example, a semiconductor memory) capable of holding the extracted feature amounts and data necessary for the operation, image data read from and written to the storage medium, and operation or personal information for the data retrieved and read from the database system. A central processing unit and an operating system capable of executing a program capable of comparing the characteristic amount and the characteristic amount extracted from the image data; and an interface device capable of communicating with external devices and transmitting and receiving data.

[0005]

Embodiments of the present invention will be described below in detail with reference to the drawings. FIG. 1 is a schematic configuration diagram of a processing system to which a method of using a feature amount for image identification according to the present invention is applied. A card 10 having a memory chip or an IC chip 4 in which personal information is converted into digital data and stored.
Is printed with a face photograph image 3 obtained by photographing the face of the card owner with a digital camera or the like. The authentication system 11 includes a computing device 1 such as a PC (personal computer) for performing image processing and data search, and a database system 2 in which personal information and feature amounts of face images are registered and recorded.
Are connected via a local bus or a network I / F card. In addition, a card reader / scanner (not shown) that reads personal information and a face image
Are connected via an appropriate I / F. The card holder sets the card reader / scanner attached to the authentication system 11 at the time of security check at the time of purchasing or entering / leaving an article. The card reader / scanner reads the coded personal information recorded on the card and simultaneously reads the face image printed on the card surface. Code information and image information are P
It is temporarily stored in an internal memory (not shown) of C1. On the other hand, the PC 1 searches for information registered in the database system 2 based on the key information included in the code information,
Receive the corresponding registration code information. Next, the registration code information contains the feature amount of the face image photographed at the time of the first registration, and is stored on the card previously obtained by reading the scanner stored in the internal memory. By comparing with the feature amount of the face image and seeing the shift amount, it is determined whether or not the card recording the personal information is that of the registered person.

(Extraction of Feature Amount) The extraction of feature amount is performed through the following two steps. (1) Preprocessing A reference point is determined from the read face image. In an embodiment, at the time of card creation, the owner is asked to face front in front of a blue background. Lighting should be illuminated from slightly above the front. An image captured by a color digital camera or an image obtained by reading a photographic image captured by a normal silver halide photography method with a color scanner or the like is used. First, the background is cut out by performing appropriate threshold processing. More specifically, since the image is taken in color, attention is paid to blue plane data in the image data, first, smoothing is performed by a median filter, and then a density value histogram is created. The thresholds x1 and x2 for extracting the background from the density value histogram are obtained from the following equation [1]. Here, h [i + 1] represents the number of pixels of the density value i. In the following equation [2], the minimum value of i that satisfies the equation [1] is x1, and in the following equation [3], the maximum value of i that satisfies the equation [1] is x2. Pixels are cut assuming that they are backgrounds. h [i] / m × 100 ≧ 1 [1] m = h [max] + h [max−1] + ·· + h [i + 1] ... [2] m = h [max] + h [max + 1] + .. + h [i-1] [3]

FIG. 6 is a diagram of an image and a background subjected to various processes, and FIG. 2A is a diagram showing an example of a histogram of the image of FIG. FIG. 7 is a diagram showing a result of processing of the skin color region, and FIG. 2B is a diagram showing an example of a histogram of the image of FIG. A histogram for the image obtained by performing the threshold processing of FIG. 6A is shown by a characteristic diagram shown in FIG. Since the area between x1 and x2 in FIG. 2A is the background, it is cut. FIG. 6E shows the background portion finally obtained by removing the noise component. Next, skin color is extracted. A histogram of green plane data is created for an area outside the background extracted by the above processing.
Here, the high density area and the low density area are cut in advance. Under the assumption that the portion occupying the largest area in the remaining image region obtained by cutting out the background region is a skin color, a skin color region is extracted in the same manner as the above-described threshold processing.
FIG. 7C is a diagram showing a result of performing the threshold processing, and FIG. 2B is an example of a corresponding histogram. A face area occupying a certain area or more and having a center of gravity close to the center of the image is defined as a face area. FIG. 7D is a diagram extracted as a face area. Further, in order to eliminate the influence of the outline, an area that does not include the outline is extracted and set as a face area (see FIG. 7E).

Next, a reference point is determined. FIG. 8 is a diagram showing a process of detecting a reference point, and FIG. 9 is a diagram showing an example of approximation on a plane. First, the extracted face area is approximated to an elliptical figure, and the density value of the green plane is inserted into a portion that is not a skin color area included therein (see FIG. 8C). From the result, the center of gravity of the eye region is obtained by applying a characteristic condition to the human eye, and the midpoint between the centers of gravity is set as a reference point (FIG. 8).
(D)). The conditions that are characteristic of the human eye here are:
Although it can be changed depending on the situation, here, the extraction is performed using the following conditions. Regions that appear to be eyes are on both sides of a perpendicular passing through the center of gravity of the face region. The region considered to be eyes is above the center of gravity of the face region. The horizontal position and area of the center of gravity of the two regions considered to be eyes are substantially equal. The variation of the density value in the region considered to be the eye is large. In addition, when multiple detections are made, the upper set is adopted as much as possible, or when an area that satisfies the conditions cannot be detected, upper and lower limits are set for each detected area. A condition for searching for a corresponding area is set, an eye area is detected, and a reference point is determined.

(2) Normal Vector FIG. 5 is a diagram showing normal vectors. The density value of the blue plane is put in the face area, and the face area is divided into small sections. For example, when an area of 400 × 400 pixels in the face area is adopted, the face area is divided into small sections of about 15 × 15 pixels. In this case, 26 × 26 = 676 small sections are obtained. Next, a plane is obtained by approximating the density value curved surface included in each small section with a least square error. The obtained orientation of the plane, that is, the normal vector of the plane is set as a feature quantity expressing the density value curve included in the small section. The normal vector of each subsection indicates a change in the shape and color of a specific area of the face area (see FIG. 5). Compared to the density value information included in this area, it can be reduced to about 1% or less.

FIG. 10 is a characteristic diagram showing a histogram of the average cosine. With respect to the normal vector extracted by the above procedure, normal vector data taken from a face image printed on a card in advance is stored in the personal information registered in the database.
The normal vector data on the database also stores the position information for the reference point. As the simplest positional information, there is a method of assigning a number to the normal vector data according to the positional relationship between the small section and the reference point. When a card is inserted into a card reader for personal authentication, a face photograph image on the card is read by a scanner incorporated in the card reader. From the read face image data, the feature amount is extracted by the PC 1 by the method described above. The normal vector data, which is the extracted feature quantity, also has position information with respect to the reference point on the image, and thus the numbers are assigned in the order of predetermined small sections. First, the database system searches the personal information obtained by the card reader for whether or not the corresponding data is registered. At this time, the search method, the search key, the search information, and the like may be arbitrarily set for each authentication system, and thus are omitted here.
If there is no corresponding information as a result of the search, it is rejected at this point due to unauthorized use or card damage. If there is the corresponding personal information, the feature amount data of the registered face image is read. In this case, it becomes normal vector data and its position information. Next, the registered normal vector data and normal vectors read and extracted from the card are sequentially compared along the position information. As a comparison method at this time, a cosine representing an angle between normal vectors corresponding to a relative position from each reference point is calculated.
That is, the deviation of the feature amount is expressed by the difference in the direction of the normal vector. In this manner, the cosine data of the normal vector representing the difference in the density value inclination of the corresponding small section plane by the number of the small sections is obtained. In the end, the cosine data is a change in the shape and color of the face area of the registered face photograph image taken in advance at the time of card creation, and the face area of the face photograph image printed on the card read by a scanner or the like at the time of authentication. And changes in shape and color. If the images are exactly the same and the conditions are the same, the directions of the respective normal vectors match, so that the cosine data becomes 1. The different the images, the smaller the cosine. For numbers that compare the cosine of the normal vector at the corresponding position between the images,
FIG. 1 is a histogram obtained by taking the averaged cosine (FIG. 1)
0).

Usually, the average cosine lies between 0 and 1. This histogram prepares a plurality of sets of face images obtained by shooting different people, and calculates a set of images obtained by changing the position, angle, etc. of the same person and calculating the average cosine between the face images. Similarly, a plurality is prepared, and the average cosine between the face images is calculated. As can be seen from this histogram, the average cosine between images of different persons first taken is always less than one. Also, the conditions for the average cosine between images of the same person are different, so
Again smaller than one. By setting and separating an appropriate threshold value 30 from these distributions, a person different from the registered individual can be photographed, or a face photograph of the same person but different from the registered face image can be obtained. It can be determined that there is. Further, since the determination can be made without using only the strict alignment, the fixed image information, and the use of only the local image information, it is possible to make a determination with extremely high robustness. That is, even if a part of the image is dirty, it is possible to identify a person without worrying about the difference or accuracy of the reading system.

The method of extracting and using the normal vector, which is a feature quantity extracted from the cut-out area, has been described in the simplest form. Here, a more effective usage will be supplemented. FIG. 3A shows a state in which a certain region is cut out from the target image 1 (region 2), and an image region that is a candidate for calculation is selected from the region, and the image region is divided into small rectangular blocks ( Region 3) is shown. How to choose area 3,
Although the division of the small section of the area 3 may be anything, a rectangular section will be described here. FIG. 3B is a diagram showing another cutout from the target image (region 1). 4 (a) and 4 (b) show the same region 3 shown in FIG. 3 (a). Here, the region 3 is rectangular as described above, and the inside is divided into small sections. The filled small sections indicate that they are actually used for calculating the feature quantity, and the blank small sections indicate that they do not contribute to the calculation of the feature quantity. Similarly, the method of selecting small sections to be used for calculation from the area 3 is 2 N −1, where N is the total number of small sections.
However, two cases will be described here as examples. The calculation formula of the feature amount can be expressed as follows. The feature amount in the case of FIG. 4A is D1, and the feature amount in the case of FIG. 4B is D2. D1 and D2 generally have different values because the small sections used for calculation in FIGS. 4A and 4B are different. Thus, different feature amounts can be set from the same image, but are assigned to different verifiers. For example, if the feature value D1 is assigned as a bank and the feature value D2 is assigned as a feature value for verification at a post office, the subject holds one image information and the bank and the post office have independent features that the bank and the post office cannot know each other. Verification can be performed by the amount. Similarly, the verifier can be fixed at one, and can select either D1 or D2 every time verification is performed. Thus, a different inspection method can be set and selected each time. D1 and D2 and / or the result of using them in combination cause a difference in the amount of calculation and the verification accuracy. This makes it possible to make a selection according to the calculation time and verification accuracy required at the time of verification.

FIG. 4A is an image of the area 3 to be calculated in FIG. 1, and the divided small sections are numbered A ₁₁ A ₁₂ ... The whole is represented by I. The small section Ik of a certain area 3 is as follows: Ik = {K ₁₁ k, K ₁₂ k · Kijk} (4) FIG. calculation which was weighted (Kij), divided subsection a _11, a _12, the weight of the corresponding portion of ·· Aij K _11, K _12, shown in · · Kij, one of the combinations Pk is: Pk = {K ₁₁ k, K ₁₂ k ·· Kijk} (5) Assuming that the calculation formula of the feature quantity according to the present invention is F, Can be expressed as F (A ₁₁ k * K ₁₁ k, A ₁₂ k * K ₁₂ k,... Aijk * Kijk) [6] = F (Ik, Pk) [7] Here, if the value of Kij or Kijk is 0, it indicates that the corresponding sub-partition image Aijk is not used for calculating the feature amount, and if it has a value of 1, FIG.
(A), and corresponds to the filled small section in FIG. 4 (b). Kij or Kijk can take one or more values. 4 (a) and 4 (b), assuming that the region 3 to be calculated is I, FIG. 4 (a) D1 = I * P1 [8] FIG. 4 (b) D2 = I * P2

[9] P ₁ , P ₂ ,... Pk are a kind of digital filter. The above equation [8],

As is clear from [9], different results D1 and D2 are obtained by changing the method P1 and P2 of extracting feature values from the same image, but Pk can be selected according to the accuracy of detection and the characteristics of the image. is there. FIG. 4 (b)
Does not use the first row part of the subdivision at all,
For example, when noise is likely to enter the first row of the small section due to the characteristics of the reading device, obtaining the feature amount from the small section image of FIG. In this way, not only is it possible to cut out an image area suitable for image characteristics such as the capability of the reading device, characteristics, color shading, shadowing, etc.
It is possible to perform a finer or more optimal feature amount comparison by appropriately selecting a use method of the small section of the cut-out region. An example in which this is applied to authentication will be described. Choose the post office as the verifier to authenticate. The post office in advance notify the P _1. Subjects associated with the ID for identifying the subject is calculated in advance D ₁ to the tag, it is registered to the post office. Storing the ID associated with the D ₁ to the IC card or the like, you print an image I to be on the card surface.
The post office reads the image I on the IC card presented by the card holder, and calculates the characteristic amount D ₁ ′ using P ₁ .
From the calculated D ₁ ′ and the ID read out from the IC card through a predetermined procedure, a pre-registered D ₁ is searched and compared, and image identification is performed. The bank advance notification of P _2, by registering the D ₂ to the bank, it is possible similar validation. In this way, it is possible to set a plurality of verifiers based on the mutually unknown feature amounts. FIG. 4E shows another embodiment of the small section. In this way, a section obtained by dividing a hollow circle into a straight line toward a center and a concentric circle may be used.

FIG. 11 is a processing flowchart of the authentication system according to the present invention. When the system starts, initialization is first performed. There are initialization of internal memory and peripheral devices, reading of setting status, connection with a database system, and the like (S4-1). When the initialization is completed, it waits for the card 10 to be inserted into the card reader (S4-
2). When the card is inserted, the personal information recorded on the card is read (S4-3). Next, the face image printed on the card is read by the image scanner (S4).
-4). When the reading of the information is completed, the personal information is searched based on the personal information (S4-5). The search key, the processing for the database system, the search method, and the like may be freely selected according to the authentication system. As a result of the search, if there is no corresponding person, the card is regarded as unauthorized use or damage to the card, etc., and rejection processing shown in FIG. After the ejection, the process returns to S4-2 again and waits for the card to be inserted. If there is such a person, the personal information and the feature amount used in the present invention are read from the registration data (S
4-7). Next, feature values are extracted from the face image data read when the card is inserted. The extracted feature amount is compared with the registered feature amount to detect a shift amount (S4-
9). If the shift amounts can be regarded as being in the same range, the authentication system performs main processing by regarding the card as a card (person identification) on which the face image of the person is printed (S4-11). The main processing includes selling an article as a credit card, releasing a lock in a security system, and the like. When the main processing is completed, the card is ejected, and the card waits for insertion again.

FIG. 12 is a flowchart showing the processing on the card creation system side for creating and registering a card.
Also in FIG. 12, when the system is started up as in FIG. 11, initialization is first performed. There are initialization of the internal memory and peripheral devices, reading of the setting state, connection to the database system, and the like (S5-1). When the initialization is completed, a request for a registration process is awaited (S5-2). When registration processing is requested by the operator, reading or input of personal information is performed (S5-3). A paper or the like in which personal information is previously written is read by a scanner or the like, and OCR processing is performed, or information is directly input by an operator. Next, the face image of the person to be registered is read or input. Here, processing such as taking a picture with a digital camera or the like and directly inputting it, or reading a photographic image taken as a normal silver halide photograph with a scanner or the like is performed (S5-4). When the face image data is captured, the feature amount is extracted next. Then, personal information is registered. The feature amounts extracted earlier are also registered in association with the personal information.
Next, memorize the registered personal information on the issuing card,
A face image is printed on the surface. The processing so far is S
5-5-8. When all processes are completed, a card is issued (S5-9).

FIG. 14 is a processing flowchart in the case where the shift amounts described in FIG. 11 cannot be regarded as being in the same range.
The authentication system always has a display device for presenting information to the operator, and displays a determination result. Of course, a sound or the like may be used instead of a visual display. Here, as the determination result on the display device of the PC,
For example, "〇〇% match. Is displayed and the card is ejected. As a result, it is known that the person who operates the authentication system may not be the person's face image, so that the face image and the person are compared and seen, or some other information (for example, a driver's license, etc.) For example, a method of making a determination by utilizing the information can also be adopted.

FIG. 15 is a flowchart of a process for extracting a feature, FIG. 16 is a flowchart of a process for determining a reference point, FIG. 17 is a flowchart of a process for obtaining a normal vector value, and FIG. It is a flowchart of the detection processing of the deviation | shift amount with a registration value. First, normal vectors are extracted after preprocessing is performed. The pre-processing is determination of a reference point. As described above, the reference point is determined by cutting out the background and extracting the face area. On the other hand, extraction of the normal vector is performed by a method as shown in FIG. The face area is divided into small sections, the plane for each section is determined, and the inclination =
Get the normal vector value. Next, detection of a deviation between the extracted feature amount and the registered feature amount is performed by a method as shown in FIG. The normal vectors at the corresponding positions are compared, and each shift amount = cosine is calculated, and the average cosine is obtained. Threshold processing is performed on the obtained average cosine.

Each step described in FIGS. 11 to 18 is converted into a program, and the program is stored in a recording medium such as a CD-ROM. It is possible to realize.

As an effect of each claim of the present invention, the input image information is subjected to arithmetic processing, and is replaced with a feature amount expressing a density value gradient accompanying a change in the object shape and color of the image. The amount of data for identifying an image can be reduced by using the amount of deviation for image identification, and adaptive processing can be performed. The feature amount extracted from the input image information is a normal vector in a density value space holding relative position information from one or more reference points on the image, and the deviation amount from the feature amount is identified. The three-dimensional shape characteristic of the object in the image can be replaced with a small amount of information by the angle formed between the normal vectors of the corresponding relative positions between the images. The amount can be reduced. As a result, it is possible to express a difference between image images with a small amount of information. In extracting feature amounts from the input image information, a plurality of characteristic shapes in the image information are determined in advance, a reference point is determined from a positional relationship between the determined shapes, and a position determination is performed based on the reference points. By taking out the set feature amount, the characteristic shape of the image is not directly used as the reference point, so that the influence of the input method, the state, and the like can be reduced. In addition, the influence of positional displacement between images is reduced. In extracting feature amounts from input image information, when extracting all or a part of the original image using the reference point and performing image identification determination using deviation of the feature amount extracted therefrom By using all or a part of the sampled feature values as needed, by using a part of the image, the unused part can be used for other purposes (for example, digital watermarking). Becomes
In addition, when the computational power is high, very high-density reading and extraction of the feature amount can be performed to perform a high-precision comparative identification process. By roughly extracting the amount, comparison using a part of the characteristic amount originally possessed becomes possible.

In extracting the characteristic amount from the input image information, an area for calculating the characteristic amount from the clipped image is specified, and / or a characteristic amount calculation method is specified to obtain the characteristic amount, and the characteristic amount is obtained. By using the same image, comparison using different feature amounts for each verifier or each verification opportunity can be performed using the same image. When inputting parameters in the calculation method, it is possible to perform a feature amount calculation with a small amount of calculation for performing a more accurate verification comparison by reflecting the previous result. Further, when the accuracy of the image read by the detection system deteriorates only in a specific region, it is possible to calculate the feature amount by excluding in such a region in advance. Instead, it became possible to secure reliability against system degradation. Whether the input image is obtained by photographing a person's face, calculates a shift amount between the extracted feature amount and the previously extracted and held feature amount, and is the same as the image from which the feature amount is extracted in advance. By judging whether or not the image is applied to a face image, it is often taken on a daily basis as compared with a fingerprint or the like, and open information can be used. It can be visually distinguished, and it is considered that a face image will be frequently used as information to be printed on a card or the like in the future, and can be effectively used. When the input image is obtained by capturing a face of a person, and the characteristic shape for determining the reference point used in extracting the feature amount is the eyes of the person, the face image is most captured. Since the target portion is the eye region, there is no possibility that there is no information on the eye region. Further, since the eye itself is not used as the reference point, the error of the reference point can be reduced even if the information of the eye shape is partially missing.

Next, embodiments of the present invention will be described. (A) The chromaticity value of the memory color is used for background extraction and skin color region extraction. (B) Similarly, when a certain area is designated, a similar color including that color can be extracted as a background or skin color. (C) As a density value for obtaining a normal vector, not only B alone, but also all of R, G, and B may be used, or R and G may be used. (D) Even if the database has normal vector data captured at a high resolution, only a part of the data may be actually compared. Also, adaptive processing is possible by increasing or decreasing the corresponding data depending on the performance of the PC performing the processing. (E) Normalization of the face size and elimination of unnecessary components (for example, hair) can be utilized more accurately and for individual identification. (F) Since the image area not used for the determination of the normal vector has no influence even if the data changes, it is possible to carry other information. (G) The determination based on the threshold can be made by applying fuzzy logic, neurology, or the like. (H) Even with a system with very low performance, a probabilistic determination (for example, the probability that the face image of the subject is 7
0%), and finally, a system that can be determined by humans can be used, and a flexible response is possible. (I) The normal vector at a position symmetrical from the center line of the face, based on the assumption that the comparison area is limited or weighted, and that the face is almost symmetrical, rather than the comparison with the entire mean cosine. It is also possible to adopt a method of checking the difference between the deviation amounts.

[0022]

As described above, according to the present invention, in a security system such as entry / exit management or credit card payment, a face image printed on a card is used to identify whether or not the person is a registrant. In this case, the amount of data for identifying an image can be reduced. Also, since the three-dimensional shape characteristics of the object in the image can be replaced with a small amount of information, the difference between the images can be expressed with a small amount of information, and the characteristic shape of the image is not directly used as the reference point. And the effect of the state or the like can be reduced, and the influence of the positional deviation between images can be reduced. Furthermore, by using a part of the image, the part not used can be used for another purpose (such as a digital watermark). Furthermore, when the computational power is high, it is possible to perform highly accurate comparison and identification processing by performing very high-density reading and extracting feature amounts, while it is desired to have only inferior power or to increase the speed. In such a case, the feature amount can be roughly extracted, and the comparison can be performed using a part of the original feature amount.

[Brief description of the drawings]

FIG. 1 is a schematic diagram of a personal information card and an authentication system showing one embodiment of the present invention.

FIG. 2 is a diagram illustrating an example of a histogram of a B plane, and a diagram illustrating an example of a histogram of a G plane.

FIG. 3 is an explanatory diagram of another cutout from a target image.

FIG. 4 is a diagram showing an example of designation of an image area to be calculated and small sections.

FIG. 5 is an explanatory diagram of a normal vector in the present invention.

FIG. 6 is a diagram showing a process of extracting a background of an image according to the present invention.

FIG. 7 is a diagram illustrating a process of extracting a flesh-color area of an image according to the present invention.

FIG. 8 is a diagram illustrating a process of detecting a reference point of an image according to the present invention.

FIG. 9 is a diagram showing an example of approximation on a plane.

FIG. 10 is a characteristic diagram showing a histogram of an average cosine.

FIG. 11 is a processing flowchart of the authentication system according to the embodiment of the present invention.

FIG. 12 is a processing flowchart of a card creation system according to an embodiment of the present invention.

FIG. 13 is a flowchart of a reject process according to the present invention.

14 is a processing flowchart in a case where the shift amounts in FIG. 11 cannot be regarded as being in the same range.

FIG. 15 is a flowchart of a feature amount extraction process according to the present invention.

FIG. 16 is a flowchart of a reference point determination process which is a pre-process in the present invention.

FIG. 17 is a flowchart of normal vector extraction processing according to the present invention.

FIG. 18 is a flowchart of a process for detecting a deviation between an extracted feature amount and a registered feature amount according to the present invention.

[Explanation of symbols]

1 ... image processing PC, 2 ... database, 3 ... face photograph, 4
... Memory for storing information on feature amounts and identification functions, 10... Personal information storage card, 11... Authentication system.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masao Kasuga 2753 Ishii-cho, Utsunomiya-shi, Tochigi Utsunomiya University Faculty of Engineering and Information Engineering (72) Inventor Hiroharu Yoshikawa (72) Inventor Tetsuya Iida 3-12-1 Kachidoki, Chuo-ku, Tokyo Forefront Tower Ricoh System Development Co., Ltd.

Claims (8)

[Claims] 1. An image processing apparatus according to claim 1, further comprising: performing arithmetic processing on input image information; replacing the image information with a feature amount expressing a density value gradient accompanying a change in an object shape and a color of the image; and using a shift amount from the feature amount for image identification. A method of using a feature amount for image identification characterized by the following. 2. A feature value extracted from the input image information is a normal vector in a density value space holding relative position information from one or more reference points on the image. The method according to claim 1, wherein the shift amount is an angle between normal vectors of corresponding relative positions between the images to be identified. 3. In extracting a feature amount from the input image information, a plurality of characteristic shapes in the image information are determined in advance, and a reference point is determined from a positional relationship between the plurality of determined shapes. 2. The method according to claim 1, wherein the feature amount determined based on the point is extracted. 4. A feature amount is extracted from the input image information by cutting out all or a part of an original image using a reference point, and performing image identification using a shift of the feature amount extracted therefrom. The method according to any one of claims 1 to 3, wherein all or a part of the sampled feature amounts are used as necessary when the determination is made. How to use features. 5. Extracting a feature amount from the input image information, designating a region for calculating a feature amount from the cut-out image and / or specifying a feature amount calculation method to obtain a feature amount; The method according to any one of claims 1 to 4, wherein the method is used for image identification. 6. The input image is obtained by photographing a face of a person, and calculates a shift amount between a feature amount extracted and a feature amount extracted and held in advance, and the feature amount is extracted in advance. The method according to any one of claims 1 to 5, wherein it is determined whether the image is the same as the image. 7. The method according to claim 1, wherein the input image is obtained by photographing a face of a person, and a characteristic shape for determining a reference point used in extracting a feature amount is eyes of the person. Item 7. A method of using a feature amount for image identification according to any one of Items 1 to 5. 8. A recording method comprising: converting a feature amount for image identification and a method of using the feature amount according to any one of claims 1 to 7 into a program and storing the program. Medium.