JP2006252272A - Image identification device, image identification processing program, image identification processing method, and data structure of registerd image data - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image identification device capable of surely identifying only images of the same kind by matching determination even in such a case that registered image data includes large quantities of image data extremely similar to image data to be identified. <P>SOLUTION: When it is determined that the number n2 of registered images GOi of identification candidates is not narrowed to 1 in first (common condition) pattern matching processing for calculating a similarity based on comparative differences of relative position between 7 reference mask positions P1-P7 set in each registered image G01 and maximum correlative positions Q1-Q7 detected in a collation image Gj in accordance therewith, second (individual condition) pattern matching processing is executed for calculating the similarity based on comparative differences of relative position between two image parts K1 and K2 set in the registered image G0i and two maximum correlative parts I1 and I2 detected in the collation image Gj in accordance therewith. When similarity to individual identification condition information of the registered image G0i is determined, this registered image is matching-determined (identified) as an image of the same kind as the collation image Gj. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to an image identification device for identifying image data, an image identification processing program, an image identification processing method, and a data structure of registered image data.

  Conventionally, for example, a fingerprint image identification device or a face image identification device for identifying and authenticating an individual is collated with newly input image data against previously registered image data, and matches according to the degree of similarity. / Image discriminating apparatuses that determine non-coincidence and identify whether registered image data and input image data are the same have been put into practical use.

  Recently, for example, as a drug management system in a medical facility, a desired tablet sheet is identified from a wide variety of tablet sheets having a large number of similar ones. Is considered.

  The image matching methods in the conventional image identification apparatus are roughly classified into a feature point (quantity) extraction method and a pattern matching method.

  In the feature point (amount) extraction method, a plurality of points or amounts serving as features on the image data are extracted, and match / non-match is determined from the similarity of the feature points (amount).

  In the pattern matching method, image data itself (referred to as “image bit data” in order to distinguish from the entire data structure here) is compared in whole or in part, and the similarity is calculated to match / not match. Determine match.

  In the feature point (amount) extraction method, the feature point (amount) is extracted as a feature point (amount) included in the image data, and the degree of similarity is seen to extract the feature point (amount). You can't see even more detailed similarities.

  On the other hand, in the pattern matching method, since it is a comparison and collation of the image data itself (image bit data), it is possible to see detailed similarities, but on the other hand, since the collation process takes time, a large number of image data to be collated If the image collation at 1: 1 is performed for each of these, the efficiency may be low and there may be a practical problem.

  For this reason, the conventional image identification apparatus uses the feature point (quantity) extraction method and the pattern matching method together, particularly when there are a large number of image data to be collated. After narrowing down image data with a high degree of similarity, efficient processing is realized by performing image matching using a pattern matching method. For example, there is an image collation apparatus based on this pattern matching method that has been filed and published by the same applicant (see Patent Document 1).

  FIG. 12 is a schematic flowchart showing a conventional image matching and identifying process using both the feature point (quantity) extraction method and the pattern matching method.

  First, image bit data of image data to be collated is captured (step M1), and feature points (amounts) of the image are extracted from the captured collation image (step M2), and each of N image data registered in advance is obtained. The registered image data having a feature point (amount) having a high degree of similarity is narrowed down from the feature points (amount) possessed (step M3).

  Then, for each of the registered image data after narrowing down by the feature point (amount) extraction method, image matching is performed using a pattern matching method with the matching image data (step MX).

  FIG. 13 is a flowchart showing the pattern matching collation process of image data described in Patent Document 1.

  FIG. 14 shows the reference mask Pi (i = 1, 2,..., N) on the registered image G0 and the verification image G1 set in accordance with the pattern matching verification process of the image data described in Patent Document 1. It is a figure (example of n = 2) which contrasts and shows maximum correlation position (area | region) Qi (i = 1, 2, ..., n). Here, an example in which collation is performed using a luminance image is shown.

  First, an image G1 to be collated (see FIGS. 14B1 and B2) is input and stored (step X1), and then moved onto a pre-registered image G0 (see FIGS. 14A1 and 14A2). A plurality of rectangular reference masks P1 and P2 are arranged (step X2).

  Then, the maximum correlation positions (regions) Q1 and Q2 where the maximum correlation coefficient is calculated on the collation image G1 are sequentially detected for the images in the masks of the reference masks P1 and P2, respectively (steps X3 and X4). The relative positional relationship between the reference masks P1 and P2 is compared with the relative positional relationship between the maximum correlation positions (regions) Q1 and Q2 (step X5).

  The degree of similarity of the entire image is calculated according to the comparison difference between the relative positional relationship between the reference masks P1 and P2 in the registered image G0 and the relative positional relationship between the maximum correlation positions (regions) Q1 and Q2 in the verification image G1. (Step X6) If the calculated similarity is equal to or greater than a preset threshold value, it is determined that the verification image G1 matches the registered image G0 (Step X7 → X8). It is determined that the verification image G1 does not match the registered image G0 (step X7 → X9).

  In such a method described in Patent Document 1, predetermined default mask parameters (shape, size, number, arrangement position, etc.) and determination conditions (similarity or distance calculation method and determination based thereon) Only one threshold value may be stored separately from the individual image data. Here, the identification condition information used in common for these image databases is referred to as common identification condition information. The common identification condition information is shared so that identification can be performed with average accuracy for all image data. If the collation identification is performed using the common identification condition information, the collation identification can be realized relatively easily.

In the case of the method described in Patent Document 1, contrary to the above, the reference masks P1, P2,..., Pn are arranged on the collation image G1, and the maximum correlation position (region) from the registration image G0. Q1, Q2,..., Qn can be detected and collated.
JP 2000-194862 A

  However, even when the image matching process is performed using the pattern matching method as described above, if there is a plurality of image data with extremely high similarity as the images to be verified, one of these registered image data is verified. In some cases, the image data cannot be identified by being matched.

  FIG. 15 is a diagram illustrating an identification example in a case where image data that is difficult to identify even by the image pattern matching matching process described in Patent Document 1 is included.

  For example, as shown in FIG. 15, most of the characteristic images (shape, name characters, character arrangement, illustrations, manufacturer marks, etc.) are the same, such as a tablet sheet for packing a predetermined number of medicine tablets. When there are a plurality of weight type tablet sheets (see FIG. 15 (B1) and (B2)) that differ only in the character notation (weight value), the predetermined common mask parameters and determination conditions are as follows: The correlation of the image of the portion excluding only the portion of the character notation (weight value) becomes dominant, and there is little comparison difference between the relative positional relationship between the reference masks Pi and the relative positional relationship between the maximum correlation positions (regions) Qi. Since the similarity is determined to be high and the match is determined, it may be erroneously determined that the tablet sheet is the same type as the tablet sheet of image data registered in advance.

On the collation image Gj of “Cassiopon tablet 5” shown in FIG. 15 (B), a plurality of reference masks, for example, seven of P1 to P7 are arranged and set, and “Cassiopon tablet 5” shown in FIGS. 15 (A1) to (A3). The maximum correlation positions (regions) Q1 to Q7 are detected in accordance with each of “Cassiopon Tablets 10” and “Exciline Tablets 1 mg”. (See FIGS. 16 to 18)
Then, as shown in FIG. 18, with respect to the relative positional relationship between the reference masks P1 to P7 in the tablet sheet image “Cassiopon tablet 5” shown in FIG. 15 (B), clearly different tablets shown in FIG. 15 (A3). For the sheet image “Exciline Tablets 1 mg”, the relative correlation between the maximum correlation positions (regions) Q1 to Q7 is also clearly different, so the similarity is “0”, which is a value sufficiently smaller than the threshold value, and does not match As shown in FIG. 17, the similar and partially different tablet sheet image “Cassiopon Tablet 10” shown in FIG. 15A is determined between the maximum correlation positions (regions) Q1 to Q7. The relative positional relationship is not sufficiently different, and as shown in FIG. 16, it is sufficiently larger than the same threshold as the collation judgment in the same kind of tablet sheet image “Cassiopon tablet 5” shown in FIG. In a similarity "6", and there is a problem that the determination matched (homologous) incorrectly.

  The present invention has been made in view of such problems, and even when registered image data contains a lot of image data that is very similar to image data to be identified, erroneous match determination is not performed, and memory saving is achieved. Provides an image identification device, an image identification processing program, an image identification processing method, and a data structure of registered image data that can reliably and efficiently identify and identify only the same type of image data in a short time The purpose is to do.

The image identification apparatus (image identification processing program) (image identification processing method) according to claim 1 (invention 9) of the present invention compares the verification image data with a plurality of registered image data, and For determining identity with registered image data,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine identity with the registration image data, and the first collation determination means by the first collation determination means. Individual identification condition information individually registered in the registered image data determined to have identity when it is determined that the verification image data is identical to a plurality of the registered image data And a second collation determination unit that compares the collation image data with the registered image data that has been determined to be identical and determines the identity by a pattern matching method. To do.

  According to this, it is possible to determine and identify image data that can be easily identified by, for example, a first matching determination unit that is determined by a pattern matching method, and a plurality of image data remains as identification candidates by the first matching determination unit. Even in this case, based on the individual identification condition information individually registered in the registered image data determined to have the identity, the verification image data is determined to be the registered image data determined to have the identity. In comparison with the above, only the image data of the same type can be reliably determined and identified by the second collation determining means for determining the identity by the pattern matching method.

The image identification device (image identification processing program) (image identification processing method) according to claim 5 (invention 10) of the present invention compares the verification image data with a plurality of registered image data, and For determining identity with registered image data,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine identity with the registration image data, and the first collation determination means by the first collation determination means. If it is determined that the collation image data is identical to a plurality of the registered image data, individual identification condition information individually registered in the registered image data determined to be identical On the basis of the image of the collation image data, a characteristic amount peculiar to each image of the registered image data determined to be identical is newly extracted, and the collation image data has the identity. Compared with the determined registered image data, a second collation determining means for determining identity by a feature amount extraction method is provided. According to this, the image data can be easily identified. One For example, it can be determined and identified by the first collation determination unit that is determined by, for example, the pattern matching method, and even when a plurality of image data remains as identification candidates by the first collation determination unit. Based on the individual identification condition information individually registered in the registered image data, the characteristic amount specific to each image of the registered image data determined to be identical from the image of the collation image data Are extracted, and the collation image data is compared with the registered image data determined to be identical, and the second collation determination unit determines the identity by the feature amount extraction method. Only image data that is can be reliably determined and identified.

  The data structure of the registered image data according to claim 11 of the present invention is a data structure of registered image data for determining the identity with the collation image data, and is used for identification commonly used for the image database. In addition to the condition information, individual identification condition information unique to each of the plurality of registered image data is recorded.

  According to this, apart from the identification condition information commonly used for the image database, individual identification condition information can be efficiently recorded, and unique features of each of the plurality of registered image data are collated. This can be used when checking whether or not the image data exists.

  The data structure of registered image data according to claim 12 of the present invention is a data structure of registered image data for determining the identity with collation image data, and is used for identification commonly used for an image database. Individual identification is required to control whether to use one or both of collation judgment means using condition information and collation judgment means using individual identification condition information unique to each of a plurality of registered image data A sex flag is recorded.

  According to this, it is possible to efficiently use identification condition information commonly used for the image database and individual identification condition information unique to each of the plurality of registered image data.

  According to the image identification device (image identification processing program) (image identification processing method) described in claim 1 (claim 7) or claim 9 of the present invention, for image data that is easy to identify, for example, pattern matching The registered image data that are determined to be identical can be identified and identified by the first collation determining unit that is determined by the method, and the plurality of image data remains as identification candidates by the first collation determining unit. A second method for determining the identity by a pattern matching method by comparing the verification image data with the registered image data determined to be identical based on individually registered condition information for identification; The collation determination means can reliably determine and identify only the same type of image data, so that very similar image data is registered together with the image data to be identified. Even if there are more to the image data without erroneous matching determination is made, it is possible to efficiently identified.

  According to the image identification device (image identification processing program) (image identification processing method) described in claim 5 (claim 8) (claim 10) of the present invention, for image data that can be easily identified, for example, pattern matching The registered image data that are determined to be identical can be identified and identified by the first collation determining unit that is determined by the method, and the plurality of image data remains as identification candidates by the first collation determining unit. Based on the individual identification condition information registered individually, the characteristic amount specific to each image of the registered image data determined to be identical is newly extracted from the image of the collation image data. The collation image data is compared with the registered image data determined to have the sameness, and the second collation determination unit determines the identity by the feature amount extraction method. Therefore, even if there is a lot of image data that is very similar to the image data that you want to identify in the registered image data, there is no false match determination and the efficiency is improved. Can be identified.

  According to the data structure of the registered image data according to claim 11 of the present invention, separate identification condition information can be efficiently recorded separately from the identification condition information commonly used for the image database. Since it can be used to check whether or not there is a unique feature in each of the registered image data in the collation image data, only the same type of image data can be reliably and efficiently saved in a short time with memory saving. It becomes possible to identify and identify the match.

  According to the data structure of the registered image data according to the twelfth aspect of the present invention, identification condition information commonly used for the image database, individual identification condition information unique to each of the plurality of registered image data, and Therefore, it is possible to efficiently identify and identify only the same type of image data in a short time with less memory.

  Therefore, according to the present invention, even when there are many image data that are very similar to the image data to be identified in the registered image data, only the same kind of image data is saved in the memory without erroneous matching determination. It is possible to provide an image identification device, an image identification processing program, an image identification processing method, and a data structure of registered image data that can be reliably determined and identified in a short time.

  Embodiments of the present invention will be described below with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram illustrating a configuration of an electronic circuit of an image identification device according to an embodiment of the present invention.

  The image identification device includes a control unit (CPU) 11, a storage device 12, a RAM 13, an image reading device 14, a display unit 15, an input unit 16, a storage medium reading unit 18, and a transmission control unit 19. Are connected to each other.

  In response to an input signal from the input unit 16, the control unit (CPU) 11 controls a control program stored in the storage device 12 in advance, or an external storage medium 18 a such as a floppy disk FD to a storage medium such as a floppy disk drive FDD. The image identification processing program read into the storage device 12 via the reading unit 18 or the server storage device 22 of an external computer terminal (program server 21) via the communication network (Internet) 20 via the transmission control unit 19 The image identification processing program read into the storage device 12 is activated, and the operation of each part of the circuit is controlled using the RAM 13 as a work memory.

  The RAM 13 is a work memory used by the control unit (CPU) 11 when executing the above-described control program (image identification processing program).

  The image reading device 14 is a device for obtaining the registered image G0i and the collation image Gj, and is an image sensor such as an image scanner device or a CCD (charge coupled device).

  The display unit 15 is a display device, such as a CRT or a liquid crystal display, for example, that displays the determination result of the comparison between the registered image G0i and the comparison image Gj.

  The input unit 16 is an input device such as a keyboard device for the user of the identification device to instruct the control unit (CPU) 11 to issue an image acquisition instruction or start a later-described identification process.

  Next, the storage device 12 will be described in detail.

  The storage device 12 comprises a ROM of a semiconductor memory, a hard disk device HDD 12a, a RAM 12b of a semiconductor memory, or the like, and stores in advance the control program read out by the control unit (CPU) 11 immediately after the identification device is turned on. Or a communication network (Internet) 20 from an external storage medium 18a such as a floppy disk FD via a storage medium reader 18 such as a floppy disk drive FDD or from a storage device 22 of an external computer terminal (program server 21). The above-described image identification processing program unit 12 bp is read (booted) via the transmission control unit 19.

  This program unit 12bp has a reference mask area (template) P1, which is arranged on one image (G0i or Gj) to be collated in the identification apparatus, such as a reference mask position (area) corresponding to common identification condition information. Mask parameter information and determination conditions of P2,..., Pn (seven in the case of this embodiment, P1 to P7) are set and stored in advance as defaults.

  This portion is used in common for all image databases, and since it is not necessary to have each image data, a memory saving configuration can be achieved.

In the RAM 12b of the semiconductor memory, N registered image information sections 12be having the following registered image data structure prepared in advance including the registered image G0i read by the image reading device 14 are booted and stored. (Depending on the usage, the memory for registered image information may be configured in advance by a ROM or the like of a semiconductor memory.)
(Here, for the sake of simplicity, as will be described later, the tablet image is cut out from the upper part of the sheet, and the size (one of the feature values) is normalized).
FIG. 2 is a diagram showing the data structure of registered image data for one image of the registered image information section 12be stored in the RAM 12b (or ROM) of the semiconductor memory in the storage device 12 of the image identification device.

  For example, the image bit data (12be4) of the registered image G0i in the bmp format, for example, of the tablet sheet read by the image reading device 14 is the identification number (ID) data 12be1, name data 12be2, and (common) feature data 12be3. At the same time, it is stored and held as image information to be registered in the image registration process. Then, unlike the conventional data structure, individual registration condition information 12be5 peculiar to each of the registered image data is added to the one piece of registered image data.

  The feature amount data 12be3 in the registered image information is extracted based on identification condition information commonly used for the image database, and is data such as the shape, size, and color of the registered image. It is used when searching for information in this part by the point (quantity) extraction method and narrowing down.

  Further, the individual identification condition information 12be5 unique to each registered image data is an individual mask parameter of an image portion that is different from each other in a similar image and an individual determination condition using the parameter. . For example, in the registered image G0i1 of the tablet sheet “Cassiopon tablet 5” as shown in FIGS. 7A1 and 7A2, it is clear that the tablet sheet “Cassiopon tablet 10” is similar to that shown in FIGS. 8A1 and 8A2. The individual determination conditions using the relative position information between the image portion K1 different from each other and, for example, individual mask parameters arranged in the adjacent portion K2 and the maximum correlation position thereof. (The same applies to the registered image G0i2 of “Cassiopon Tablet 10”.) These are different from the mask parameters commonly used for identification with the average accuracy and the judgment conditions commonly used, and are specific to the image. Close-up local features to make them distinguishable.

  The registered image information memory 12be in the RAM 12b of the semiconductor memory in the storage device 12 of the image identification device stores N pieces of data for one image (for registered images).

  Next, the work memory RAM 13 will be described in detail.

  FIG. 3 is a diagram showing an example of contents stored in the RAM 13 of the image identification device.

  In the RAM 13, an input image memory 13a, a collation image feature amount data memory 13b, a maximum correlation position (area) memory 13c, a collation determination result memory 13d, and the like are secured.

  In the input image memory 13a, for example, an image G of a tablet sheet read by the image reading device 14 is temporarily stored as a multi-tone image (here, only luminance information).

  At the time of collation, a collation image Gj is stored. In practice, as shown in FIG. 4, noise removal, sheet shape, tablet name portion extraction, rotation of the extracted image, size standardization, and the like are performed. This will be described as a standardized image.

  Then, in the collation image feature amount data unit 13b, feature amount data that is commonly applied to the image database for the collation image is extracted according to the image identification processing program and stored in the collation image feature amount data unit 13b.

  The maximum correlation position (region) memory 13c is narrowed down from the N registered image data by the feature amount extraction method, and either one of the image data (G0i) that is a candidate and the collation image (Gj). ) The position of each maximum correlation position (region) Q1 to Q7 where the maximum correlation coefficient is calculated on the other image (Gj or G0i) with respect to each of the reference mask regions, eg, P1 to P7, arranged on the upper side (Representative coordinate) information is detected and stored.

  The collation determination result memory 13d is narrowed down from the N registered image data by the feature amount extraction method, and for each image (G0i or Gj) of each of the candidate images and the collation image (Gj). Difference between the mutual positional relationship ΔXYp of all the reference mask regions P1 to P7 and the mutual positional relationship ΔXYq of the maximum correlation positions (regions) Q1 to Q7 on the other image (Gj or G0i) (first pattern matching) ) Furthermore, the determination result of coincidence / non-coincidence between each registered image G0i and the collation image Gj, which are identification candidates to be evaluated according to second pattern matching described later using the individual identification condition information 12be5. Is memorized.

  Next, the image collating and identifying function of the first embodiment in the image identifying device having the above-described configuration will be described.

  FIG. 5 is a flowchart showing image collation identification processing (part 1) of the first embodiment in the image identification apparatus.

  FIG. 6 is a flowchart showing image collation identification processing (part 2) of the first embodiment in the image identification apparatus.

  First, the image processing program stores many common feature quantities stored in the registered image memory 12be based on the feature quantity data 13b extracted from the matching image (Gj) stored in the input image memory 13a in the RAM 13. Image matching processing (search and narrowing processing) (see FIG. 12) by the feature point (quantity) extraction method is applied to the feature quantities that have been extracted and registered in advance for the registered images G0i (i = 1, 2,..., N) The registration image data is narrowed down to a plurality of identification candidate registration images G0i (i = 1, 2,..., N1) similar to the collation image 13Gj (step S1).

  For example, a search is executed to check whether there is an item in the registered image G0i (i = 1, 2,..., N) that matches the feature amount (for example, the width of the sheet) extracted from the collation image Gj “Cassiopon tablet 5”. The tablet sheet is narrowed down to, for example, five (n1) tablet sheets similar to the “Cassiopon tablet 5”.

  Then, the reference mask positions (regions) P1 to P7, which are a plurality of common identification conditions set on the registered image G0i (or the collated image Gj), and the collated image Gj (or the registered image G0i) according to the reference mask positions (regions) P1 to P7. A first (common condition) pattern matching process (see FIG. 13) for calculating a similarity based on a relative comparison difference with a plurality of detected maximum correlation positions (regions) Q1 to Q7 extracts the feature points (quantities). Registration image G0i (i = i = 10) that is repeatedly executed with each registered image G0i (i = 1, 2,..., N1) of the identification candidates after the image matching process (step S1) by the method. It narrows down to 1,2,..., N2 <n1) (steps S2, S3).

  For example, if the unknown tablet sheet is “Cassiopon Tablet 5”, two very similar (n2) registered tablet sheets (“Cassiopon Tablet 5” and “ Casiopon lock 10 ") (see FIG. 15).

  In the first pattern matching process (steps S2 and S3), the reference mask may be arranged on either the registered image side or the collation image side.

  Then, it is judged whether or not the number n2 of registered images G0i of identification candidates narrowed down by the first pattern matching processing (steps S2 and S3) has been narrowed down to one (n2 = 1) (step S4), for example. When it is determined that the number n2 of the identification candidate images G0i is narrowed down to one (n2 = 1), the registered image G0i narrowed down to one is the same type that matches the matching image Gj. It is identified and determined that there is (step S4 → S5).

  On the other hand, the number n2 of identification candidate registration images G0i narrowed down by the first pattern matching processing (steps S2 and S3) is not narrowed down to one, and is the same as the registration image G0j of the tablet sheet “Cassiopon Tablet 5”, for example. Only two very similar tablet sheets (“Cassiopon Tablet 5” and “Cassiopon Tablet 10”) including one kind have been narrowed down. S4.fwdarw.S6), the process proceeds to the image collation identification process (part 2) in FIG.

  FIG. 7 shows individual identification condition information 12be5 on the registered image G0i1 side when the registered image G0i1 and the collated image Gj are of the same type in accordance with the image collation / identification processing (part 2) of the first embodiment in the image identification device. It is a figure which shows the pattern matching state using.

  FIG. 8 shows individual identification condition information on the registered image G0i2 side when the registered image G0i2 and the verification image Gj are similar and different types in accordance with the image verification identification process (part 2) of the first embodiment in the image identification device. It is a figure which shows the pattern matching state using 12be5.

  For example, as a registered image G0i that is an identification candidate for an unknown collating tablet sheet “Cassiopon Tablet 5” as shown in FIGS. 7B1 and 7B2, FIGS. 7A1 and 7A2. And FIGS. 8 (A1) and (A2), the images are narrowed down to registered images G0i1 and G0i2 of two very similar tablet sheets (“Cassiopon Tablet 5” and “Cassiopon Tablet 10”) containing the same species. When the process proceeds to the image collation identification process (part 2) in FIG. 6, the registered image G0i is determined based on the individual identification condition information 12be5 stored in the registered image information memory 12be in the RAM 12b. The degree of similarity is calculated based on the relative comparison difference between the set two image portions (regions) K1 and K2 and the two maximum correlation portions (regions) I1 and I2 detected on the matching image Gj accordingly. First The pattern matching process (individual condition) is sequentially performed between the registered images G0i (i = 1, 2) and Gj of the respective identification candidates after the first (common condition) pattern matching process (steps S2 and S3). A registered image G0i (i = 1, 2,..., N3 <n2) that is executed and is similar to the individual features included in the individual identification condition information of the registered image G0i as candidates of the same type of collation image data (Steps S7 and S8).

  In the second (individual condition) pattern matching process, unlike the first (common condition) pattern matching process, the reference mask is always arranged on the registered image side.

  For example, as shown in FIGS. 7 (A2) and (B2), two set image portions (areas) K1, K1 and D2 are set according to the individual identification condition information of the registered image G0i1 of the tablet sheet “Cassiopon Tablet 5” remaining as an identification candidate. The relative comparison difference between K2 and the two maximum correlation portions (regions) I1 and I2 detected on the collation image Gj of the same type of tablet sheet “Cassiopon tablet 5” is actually “0”. Alternatively, it becomes almost “0”, and the registered image G0i1 of the tablet sheet “Cassiopon Tablet 5” is judged to have a high similarity with the unknown collated tablet image Gj, while as shown in FIGS. 8A2 and 8B2. In addition, based on the individual identification condition information of the registration image G0i2 of the tablet sheet “Cassiopon Tablet 10” remaining as an identification candidate among the registration image data, the set two image portions (areas) K1, K2 and correspondingly Match Actually, the relative comparison difference between the two maximum correlation portions (regions) I1 and I2 detected on the collation image Gj of the different kind of tablet sheet “Cassiopon tablet 5” becomes large, and the tablet sheet “Cassiopon tablet 10” The registered image G0i2 has a low similarity to the unknown collated tablet image Gj, and is judged to be inconsistent.

  Then, it is determined whether or not the number n3 of registration images G0i of identification candidates narrowed down by the first and second pattern matching processes (steps S2 to S8) has been narrowed down to one (n3 = 1) (step S9). ) For example, if it is determined that the number n3 of the tablet sheet “Cassiopon tablet 5” as the candidate registration image G0i1 has been narrowed down to one (n3 = 1), the narrowed registration The tablet sheet “Cassiopon tablet 5” as the image G0i1 is identified and determined to be the same type that matches the tablet sheet “Cassiopon tablet 5” as the collation image Gj (steps S9 → S10).

On the other hand, if it is determined that the number n3 of registered images G0i as identification candidates has not been narrowed down to one even after the first and second pattern matching processes (steps S2 to S8), they are still. When it is determined that there is a mixture of different types (steps S9 → S11), the image collation / identification process with strict determination conditions by the above-described method is repeated. (Alternatively, “n3 types of image data are identified as identification candidates” as identification results.)
If the number n3 of registered images G0i as identification candidates has become 0 after the second pattern matching processing (steps S2 to S8), the collation image data is identified and determined as unregistered image data. The

  Therefore, according to the image collation / identification function of the first embodiment in the image identification apparatus having the above-described configuration, the plurality of reference mask positions (regions) P1 to P7 set on the registered image G0i, and the collation image Gn corresponding thereto. The number of identification candidate registration images G0i narrowed down by the first (common condition) pattern matching process for calculating the similarity based on the relative comparison differences with the plurality of maximum correlation positions (regions) Q1 to Q7 detected When it is determined that n2 has not been narrowed down to one, for example, two image portions set for the registered image G0i based on the individual identification condition information 12be5 stored in the registered image memory 12be. (Region) K1 and K2, and a second degree of similarity calculation based on the relative comparison difference between the two maximum correlation portions (regions) I1 and I2 detected on the matching image Gj. By executing the (individual condition) pattern matching process, the collation image Gj is an image of the same type as the registered image G0i that is similar to the individual features included in the individual identification condition information of the registered image G0i. Since the match is determined (identified), for example, even when image data of the tablet sheet “Cassiopon Tablet 10” that is very similar to the registered image data of the tablet sheet “Cassiopon Tablet 5” to be identified exists as a candidate, Thus, only the image data of the tablet sheet “Cassiopon tablet 5” of the same type can be reliably determined and identified without remaining matched.

  And this 2nd pattern matching process is not performed with respect to all the image data. In the case of image data that can be easily identified (in the case of n2 = 1), there is an advantage that the identification time is short and it is not necessary.

  Here, the determination conditions used for the second pattern matching processing are shown as being included in the individual identification condition information in the same manner as the individual mask parameters, but individual identification using the same common determination conditions. A configuration not registered in the usage condition information is also conceivable.

  In the first embodiment, when it is determined that the number n2 of registered images G0i of identification candidates narrowed down by the first (common condition) pattern matching processing is not narrowed down to one, further registration is performed. The number n3 of registered images G0i of identification candidates is the same type as the collation image Gj by the second (individual condition) pattern matching processing based on the individual identification condition information 12be5 stored in the image memory 12be. Although it is configured to narrow down to the number, as will be described in the next second embodiment, registered images of identification candidates by a series of image matching processes by a feature point (quantity) extraction method and a first (common condition) pattern matching method If the number n1 of G0i is not narrowed down to one, another feature portion is additionally extracted and image matching processing using the same feature point (quantity) extraction method is performed. It may be configured to realize narrowing down to one registered image G0i of identification candidates by additionally executing.

(Second Embodiment)
FIG. 9 is a flowchart showing image collation identification processing of the second embodiment in the image identification apparatus.

  That is, in the image collation identifying process of the second embodiment, first, it is stored in the registered image information section memory 12be in the same manner as the image collating / identifying process (part 1) in the first embodiment (see FIG. 5). The image matching process (step A1) based on the feature point (quantity) extraction method based on the feature quantity data 12be3, and a plurality of reference mask positions (regions), for example, P1 to P7 set on the registered image G0i, and correspondingly An image matching process (step A1 ′) based on a pattern matching method based on a comparison difference between the relative positional relationships between the plurality of maximum correlation positions (regions) Q1 to Q7 detected on the matching image Gj is executed. It is determined whether or not the number n2 of registered images G0i as identification candidates has been narrowed down to one (n2 = 1) (step A2).

  For example, when it is determined that the number n2 of the registered images G0i is narrowed down to one (n2 = 1), the registered image G0i narrowed down to one is the same type that matches the matching image Gj. It is discriminated and determined to be a thing (step A2 → A5).

  On the other hand, the number n2 of registration images G0i of identification candidates narrowed down by the two-stage image matching processing (steps A1 and A1 ′) is not narrowed down to one, for example, the matching image Gj of the tablet sheet “Cassiopon Tablet 5” If the registered images G0i1 and G0i2 of two very similar tablet sheets (“Cassiopon Tablet 5” and “Cassiopon Tablet 10”) including the same type are narrowed down to two or more, It is still determined that there is a mixture of different types, and a further feature portion different from the image matching process by the common feature point (amount) extraction method in step A1 is additionally extracted (step A3), and this additional feature point (amount). The image matching process by the feature point (quantity) extraction method based on the above is additionally executed (step A4).

  Here, a new feature point (amount) to be additionally extracted and a determination condition for collation are specific to the registered image data registered in advance in the individual identification condition information 12be5.

  For example, as shown in FIGS. 15 (A1) and (A2), when the registered images G0i of the identification candidates are narrowed down to two tablet sheets (“Cassiopon tablet 5” and “Cassiopon tablet 10”), the registered image memory In the registered image G0i1 of the tablet sheet “Cassiopon Tablet 5” stored in advance in 12be, yet another individual feature point (amount) data (for example, the number “5”) registered in the identification condition information 12be5 in advance. Based on the corresponding image) extraction instruction, a process for additionally extracting whether or not the same feature point (quantity) data “5” exists in the collation image Gj is performed, and a comparison collation process is performed based on the result (step A3). , A4), it is discriminated that the tablet sheet as the collation image Gj is of the same type that matches the tablet sheet “Cassiopon tablet 5” as the registered image G0j (step A5).

  On the other hand, based on an instruction to extract further individual feature point (amount) data (for example, the image corresponding to the number “10”) in the registered image G0i2 of “Cassiopon Tablet 10”, the same feature point (amount) ) When the process of additionally extracting whether or not the data “10” exists is performed and the comparison / collation process is performed based on the result, the difference is large, and it is correctly determined that the data is different.

An example of a new feature extraction is a color feature. (In the above example, since image matching is performed on a luminance image, for example, color information is first extracted here.)
As another example of feature quantity extraction, a method of finding an area corresponding to “5” and “10” by a projection process (in the horizontal direction and the vertical direction) and performing a labeling process on the area is also conceivable. Then, in the case of “5”, the number of labels (the number of two-dimensional connected components) is 1, and in the case of “10”, the number of labels is 2, so that both can be discriminated. Such an additional feature amount and determination condition may be provided on the registration side.

  Therefore, according to the image collation / identification function of the second embodiment in the image identification apparatus having the above-described configuration, the first-stage image collation processing by the feature point (quantity) extraction method and the second-stage image collation processing by the pattern matching method. If it is determined that the number n2 of registered images G0i of identification candidates narrowed down through 1 is not narrowed down to 1, the individual features existing in the registered image G0i stored in the registered image memory 12be are further determined. By performing the third stage image collation processing by the feature point (quantity) extraction method based on the point (quantity) data, the collation image Gj is similar to the individual feature points (quantity) of the registered image G0i. Since it is determined to match (identify) as the same type of image data as the registered image G0i, it is very similar to the registered image data of the tablet sheet “Cassiopon Tablet 5” to be identified, for example. Even if the image data of the tablet sheet “Cassiopon Tablet 10” exists as a candidate, the image data of only the same type of tablet sheet “Cassiopon Tablet 5” is surely kept without being erroneously determined. It can be identified by matching.

  The additional feature point (amount) extraction process is not performed on all image data. In the case of image data that can be easily identified (in the case of n2 = 1), there is an advantage that the identification time is short and it is not necessary.

  Although it is advantageous in terms of identification time that data on the registered image side of new feature points (amounts) to be additionally extracted is registered in advance in individual identification condition information 12be5 at the time of registration, From the viewpoint, the individual feature point (quantity) data itself on the registered image side can be detected at the time of collation and not registered in the individual identification condition information 12be5.

(Third embodiment)
In the image identification device according to the first embodiment, for example, an image pattern of an image portion that is individually and clearly different among similar images and corresponding to the registered pixel data 12be4 (G0i) in the registered image information section memory 12be Individual identification condition information 12be5 (see FIGS. 7A2 and 8A2) which is position information is stored, and the image matching process and the pattern matching method in the first stage by the feature point (amount) extraction method are stored. If the number of identification candidates n2 of the registered image G0i is not narrowed down to one by the image matching process in the second stage, additional image matching by the pattern matching method based on the individual identification condition information 12be5 unconditionally. The process is executed to identify the matching image Gj that finally matches the registered image G0i. The image identifying apparatus according to the third embodiment will be described below. As described above, the individual identification condition information 12be5 is not prepared for easily identifiable image data having features that are clearly different from various verification images Gj, and the additional image verification processing by the pattern matching method is not executed. It is good.

  FIG. 10 is a diagram illustrating a data structure of one image of the registered image information unit 12be stored in the RAM 12b (or ROM) of the semiconductor memory in the storage device 12 of the image identification device according to the third embodiment of the present invention. is there.

  That is, in the image identification device according to the third embodiment, as shown in FIG. 10A, for example, a registered image G0i that has no similar image data and can be easily identified needs to be individually identified. The individual identification condition information 12be5 is not prepared by setting the sex flag 12bef to “0”, and the image matching processing by the second (individual condition) pattern matching method is not performed. On the other hand, as shown in FIG. 10B, for example, for a registered image G0i that is not easily identified due to the presence of a lot of similar image data when identifying image data, the individual identification necessity flag 12bef is set to “1”. As in the first embodiment, individual identification condition information 12be5 is prepared, and image matching processing is performed by the second (individual condition) pattern matching method.

  As a result, it is not necessary to prepare individual identification condition information 12be5 for all registered images (G0i (i = 1, 2,...)). Therefore, it is possible to efficiently improve the identification accuracy of the image data while saving the memory capacity.

  Further, the flag 12bef is further expanded (using 2 bits), for example, “2” has another meaning, and n1 identification candidate flags after image collation processing by the feature point (quantity) extraction method are performed. When 12bef is searched and all are “2”, it is also possible to perform image matching using only individual identification condition information 12be5 without performing the first pattern matching process using the common identification condition information. is there. When individual features are very unique, the identification time can be shortened in the case of image data that can be identified without performing the first pattern matching process.

(Fourth embodiment)
In the image identification device according to the first embodiment, for example, in the case of the tablet sheet “Cassiopon lock 5”, corresponding to the registered pixel data 12be4 (G0i) in the registered image information section memory 12be (FIG. 7 (A1) / FIG. 8). (See (A1)), “5” image portion (region) K1 and its adjacent image portion (region) K2 that are individually and clearly different among similar images, and their mutual positional information for individual identification. The condition information 12be5 is stored in advance, but as described in the image identification apparatus of the fourth embodiment, the image portion Ki as the individual identification condition information 12be5 corresponding to the registered pixel data 12be4. It is good to set it as the structure which can be set-stored according to user operation.

  FIG. 11 is a diagram showing a user setting state of individual identification condition information 12be5 corresponding to the registered pixel data 12be4 in the image identification apparatus according to the fourth embodiment of the present invention.

  That is, when the registered image G0i is additionally stored and registered in the registered image information section memory 12be, the image G0i to be registered input by the image reading device 14 is displayed as shown in FIG. 11 (A1). For example, an image portion (region) Ki of “5”, which is individually and clearly different among similar images in the state displayed on the portion 15, is displayed on the key on the input portion 16 as shown in FIG. Mask width Wxi or mask height Hyi, mask position (, interval), etc. according to the input operation are set by user operation, mask parameters of a plurality of masks (i = 1, 2, 3,...), A determination condition or the like can be registered or corrected as individual identification condition information 12be5.

  According to this, even when there is image data to be newly registered, the individual identification condition information 12be5 corresponding to the new registered image G0i can be set and registered easily and appropriately. In addition, the user can finely adjust the identification accuracy.

  Note that each processing method by the image identification apparatus described in each of the above embodiments, that is, the image collation identification processing (part 1) of the first embodiment shown in the flowchart of FIG. 5, and the first embodiment shown in the flowchart of FIG. Each of the methods such as the image collation identification process (No. 2) and the image collation identification process of the second embodiment shown in the flowchart of FIG. 9 is a memory card (ROM card, RAM) as a program that can be executed by a computer. Card), a magnetic disk (floppy disk, hard disk, etc.), an optical disk (CD-ROM, DVD, etc.), an external recording medium 18a such as a semiconductor memory, and the like can be distributed. Various computer terminals having an image reading function read the program stored in the external recording medium 18a into the storage device 12 by the storage medium reading unit 18, and the operation is controlled by the read program. The image collation / identification function described in each of the above embodiments can be realized, and the same processing can be executed by the method described above.

  The program data for realizing each of the above methods can be transmitted on the communication network (Internet) 20 in the form of a program code, and a computer terminal (program server) connected to the communication network (Internet) 20 It is also possible to implement the above-mentioned image collation identification function by fetching the program data from the storage device 22 of 21).

  In the above-described embodiment, an example of performing collation identification using a luminance image is shown, but it may be performed for each color.

  Furthermore, although an example in which the determination is made based on the relative positional relationship of the mask having the maximum correlation has been shown, a simple determination such as “matches when the maximum correlation coefficient value of the mask of interest matches a certain threshold (for example, 0.8) or more” A condition may be applied to identify both.

  In the embodiment, the feature point (amount) extraction method and the pattern matching method are used together in the present embodiment. First, image data having a high degree of similarity is narrowed down in advance by the feature point (amount) extraction method, and then the pattern matching method is used. As shown in the example of performing image collation by, the fact that it can be applied to the case of performing identification without narrowing down by the feature point (quantity) extraction method because the registered image data is small or the device is a high-speed device. Needless to say.

  Furthermore, in the above-described embodiment, the present invention has been described as an example in which identification is performed using image data of a tablet sheet. However, it is needless to say that the present invention can be applied to other general image data.

  Note that the present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention at the stage of implementation. Furthermore, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements. For example, even if some constituent requirements are deleted from all the constituent requirements shown in each embodiment or some constituent features are combined, the problems described in the column of the problem to be solved by the invention can be solved. When the effects described in the column of the effect of the invention can be obtained, a configuration in which these constituent elements are deleted or combined can be extracted as an invention.

1 is a block diagram showing a configuration of an electronic circuit of an image identification device according to an embodiment of the present invention. The figure which shows the data structure for 1 image of the registration image information part 12be memorize | stored in RAM12b (or ROM) of the semiconductor memory in the memory | storage device 12 of the said image identification apparatus. The figure which shows the example of the content memorize | stored in RAM13 of the said image identification apparatus. The figure which shows the mode of extraction of the tablet name part of the tablet sheet image read by the image reader 14 of the said image identification apparatus. 3 is a flowchart showing image collation identification processing (part 1) of the first embodiment in the image identification apparatus. 7 is a flowchart showing image collation identification processing (part 2) of the first embodiment in the image identification device. A pattern using individual identification condition information 12be5 on the registered image G0i1 side when the registered image G0i1 and the verification image Gj are of the same type in accordance with the image verification identification process (part 2) of the first embodiment in the image identification device The figure which shows a matching state. In accordance with the image collation / identification processing (part 2) of the first embodiment in the image identification apparatus, individual identification condition information 12be5 on the registered image G0i2 side when the registered image G0i2 and the collation image Gj are similar and different types is used. The figure which shows a pattern matching state. 9 is a flowchart showing image collation identification processing according to the second embodiment in the image identification apparatus. The figure which shows the data structure for 1 image of the registration image information part 12be memorize | stored in RAM12b (or ROM) of the semiconductor memory in the memory | storage device 12 of the image identification device which concerns on 3rd Embodiment of this invention. The figure which shows the user setting state of the individual identification condition information 12be5 matched with the registration image information part 12be4 in the image identification device which concerns on 4th Embodiment of this invention. 10 is a general flow chart showing a conventional image matching and identifying process using a feature point (amount) extraction method and a pattern matching method in combination. 9 is a flowchart showing pattern matching matching processing for image data described in Patent Document 1; The reference mask Pi (i = 1, 2,..., N) on the registered image G0 set in association with the pattern matching matching process of the image data described in Patent Document 1 and the maximum correlation position ( Region) Qi (i = 1, 2,..., N) for comparison (n = 2 example). The figure which shows the example of an identification in case the image data difficult to identify according to the pattern matching collation process of the image data of the said patent document 1 are included FIG. 10 is a diagram showing a comparison between a reference mask Pi on a collation image and a maximum correlation position (region) Qi on a registered image of the same type of sheet set in association with the pattern matching collation processing of image data described in Patent Document 1; Compare the maximum correlation position (region) Qi on the registered image of a sheet that is different from the reference mask Pi on the collation image set in the pattern matching collation processing of the image data described in Patent Document 1 but has a high degree of similarity. FIG. The figure which compares and shows the reference | standard mask Pi on the collation image set with the pattern matching collation process of the image data of the said patent document 1, and the largest correlation position (area | region) Qi on the registration image of a dissimilar sheet.

Explanation of symbols

11 ... CPU
12: Storage device 12a: Semiconductor memory ROM and hard disk drive HDD
12b ... RAM of semiconductor memory
12be1 ... identification number (ID) data 12be2 ... name data 12be3 ... (common) feature value data 12be4 ... registered image bit data 12be5 ... individual identification condition information 13bef ... individual identification necessity flag 13 ... (work memory) RAM
13a ... Input image G (collation image Gj) memory 13b ... Collation image feature data memory 13c ... Maximum correlation position (area) memory 13d ... Collation determination result memory 14 ... Image reading device 15 ... Display unit 16 ... Input unit 17 ... Bus DESCRIPTION OF SYMBOLS 18 ... Storage medium reading part 18a ... External storage medium 19 ... Electric transmission control part 20 ... Communication network (Internet)
21 ... Program server 22 ... Server storage G0i ... Registered image Gj ... Verification image

Claims (12)

An image identification device that compares matching image data with a plurality of registered image data to determine identity with the registered image data,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine identity with the registered image data;
If it is determined by the first collation determining means that the collation image data is identical to a plurality of the registered image data, the collation image data is individually registered in the registration image data determined to be identical. Second matching determination means for comparing the matching image data with the registered image data determined to have the identity based on the individual identification condition information, and for determining the identity by a pattern matching method. When,
An image identification apparatus comprising: The first collation determining means includes a plurality of criteria applied as conditions common to all collations on either one of the images of the plurality of registered image data and the image of the collation image data. First reference area setting means for setting an area;
The common condition on the one image set by the first reference area setting unit on the other image of the images of the plurality of registered image data and the image of the collation image data First maximum correlation area detecting means for detecting a plurality of reference areas and a plurality of maximum correlation areas each having a maximum correlation coefficient;
The relative positional relationship of a plurality of reference areas applied as the common condition on the one image set by the first reference area setting means and the other detected by the first maximum correlation area detection means First identity determination means for determining identity between the one image and the other image based on a difference between relative positions of a plurality of maximum correlation areas on the image of
The image identification apparatus according to claim 1, further comprising: The second collation determining means includes
Second reference area setting means for setting a plurality of individual reference areas on the images of the plurality of registered image data based on information registered in the individual identification condition information;
On the image of the collation image data, each of the plurality of individual reference areas on the respective images of the plurality of registered image data set by the second reference area setting unit has a maximum correlation coefficient. Second maximum correlation area detecting means for detecting a plurality of maximum correlation areas;
Relative positional relationship between the plurality of individual reference areas on each of the plurality of registered images set by the second reference area setting means and the collation detected by the second maximum correlation area detecting means Second identity determination means for determining identity between each of the plurality of registered images and the collation image based on a difference from a relative positional relationship between a plurality of maximum correlation areas on the image;
The image identification apparatus according to claim 1, further comprising:   The image identification apparatus according to claim 1, wherein the individual identification condition information can be set according to a user operation. An image identification device that compares matching image data with a plurality of registered image data to determine identity with the registered image data,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine identity with the registered image data;
If it is determined by the first collation determining means that the collation image data is identical to a plurality of the registered image data, the collation image data is individually registered in the registration image data determined to be identical. Based on the individual identification condition information, a characteristic amount specific to each image of the registered image data determined to be identical is newly extracted from the image of the collation image data, and the collation A second collation determination unit that compares the image data with the registered image data determined to have identity, and determines the identity by a feature amount extraction method;
An image identification apparatus comprising: Collation for narrowing down the identity of the plurality of registered image data and the collation image data by a feature amount extraction method based on a feature amount extracted in common from an image database before the first collation determination unit The image identification apparatus according to claim 1, further comprising a determination unit. An image identification processing program for controlling a computer of an image identification apparatus that compares collation image data with a plurality of registered image data and determines the identity with the registered image data,
The computer,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine the identity with the registered image data, and the first collation determination means by the first collation determination means. If it is determined that the collation image data is identical to a plurality of the registered image data, individual identification condition information individually registered in the registered image data determined to be identical A second collation determination unit that compares the collation image data with the registered image data that has been determined to be identical, and determines identity by a pattern matching method,
A computer-readable image identification processing program adapted to function as a computer program. An image identification processing program for controlling a computer of an image identification apparatus that compares collation image data with a plurality of registered image data and determines the identity with the registered image data,
The computer,
First collation determination means for comparing the collation image data with the registered image data based on common identification condition information to determine the identity with the registered image data, and the first collation determination means by the first collation determination means. If it is determined that the collation image data is identical to a plurality of the registered image data, individual identification condition information individually registered in the registered image data determined to be identical On the basis of the image of the collation image data, a characteristic amount peculiar to each image of the registered image data determined to be identical is newly extracted, and the collation image data has the identity. A second collation determination means for comparing the determined image data with the determined registered image data by a feature amount extraction method;
A computer-readable image identification processing program adapted to function as a computer program. An image identification processing method performed by a computer for comparing a plurality of registered image data stored in a memory and collation image data input by an image input device to determine the identity of the registered image data,
A first collation determination step of comparing the collation image data with the registration image data based on common identification condition information to determine identity with the registration image data;
If it is determined by the first collation determining means that the collation image data is identical to a plurality of the registered image data, the collation image data is individually registered in the registration image data determined to be identical. A second collation determination step of comparing the collation image data with the registered image data determined to be identical based on the individual identification condition information, and determining the identity by a pattern matching method. When,
An image identification processing method performed by a computer, comprising: An image identification processing method performed by a computer for comparing a plurality of registered image data stored in a memory and collation image data input by an image input device to determine identity with the registered image,
A first collation determination step of comparing the collation image data with the registered image data based on common identification condition information to determine identity with the registered image data;
When it is determined by the first collation determination means that the collation image data is identical to a plurality of the registered image data, the collation image data is individually registered in the registration image data determined to be identical. Based on the individual identification condition information, a characteristic amount specific to each image of the registered image data determined to be identical is newly extracted from the image of the collation image data, and the collation A second collation determination step of comparing the image data with the registered image data determined to be identical and determining the identity by a feature amount extraction method;
An image identification processing method performed by a computer, characterized by comprising: A data structure of registered image data for determining identity with collation image data,
A data structure of registered image data, characterized in that individual identification condition information unique to each of a plurality of registered image data is recorded separately from identification condition information commonly used for an image database. A data structure of registered image data for determining identity with collation image data,
Either one of collation determination means using identification condition information commonly used for the image database and collation determination means using individual identification condition information unique to each of a plurality of registered image data, or A data structure of registered image data in which an individual identification necessity flag for controlling whether to use both is recorded.

Images