JP2001209651A - Method and device for retrieving multi-dimensional vector and recording medium having multi-dimensional vector retrieval program recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem that processing time and a processing amount required for the processing of retrieving n-dimensional vector data increase corresponding to the dimension number of vector data to be a retrieval object. SOLUTION: In this multi-dimensional vector retrieval method for retrieving pertinent multi-dimensional vector data whose position and size are both present inside an optional multi-dimensional rectangular area inside a multi-dimensional space from the stored plural multi-dimensional vector data, the multi-dimensional rectangular area is inputted as a retrieval range, a first judgment processing is performed to a data pair composed of the multi-dimensional vector data and an address value calculated based on the rough values of the respective dimensions of the vector data by using the address value, a second judgment processing is performed by using the multi-dimensional vector data in the case that the address value is within a prescribed range and the multi-dimensional vector data of the data pair are outputted as a retrieved result in the case that the vector data are within the prescribed range.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] The present invention relates to a method for producing a plurality of n (n ≧ n).
1) From a database or the like in which the dimensional vector data is stored, an arbitrary n in both the position and the size in the n-dimensional space.
The present invention relates to a multidimensional vector search method for searching for vector data existing in a dimensional rectangular area, and more particularly to a technique that is effective when applied to a search when n is several tens or more.

[0002]

2. Description of the Related Art Heretofore, a method of searching for vector data existing in an arbitrary n-dimensional rectangular area in terms of both position and size in an n-dimensional vector space from a plurality of accumulated n-dimensional vector data has been disclosed by PCT. / EP97 / 04
520.

In this method, a vector space is divided into regions that can be ordered one-dimensionally, and the vector data is managed by the region to which each vector data belongs. At the time of search, any position and size in the vector space are arbitrary n
With the dimensional rectangular area as the search range, all the areas overlapping the search range are obtained, and for the vector data existing in each obtained area, the minimum value and the maximum value of each dimension value of the vector data and each dimension of the search range are obtained. Is compared. In comparing the respective dimension values of the vector data, the vector data determined to be within the search range is output as a search result.

[0004]

However, the conventional method as described above has a problem that when the target data is a high-dimensional vector having several tens of dimensions or more, the processing time required for searching increases. That is, according to the above-described known example, the search processing is “processing for obtaining all areas overlapping with the search range” and “processing for comparing each dimension value of the vector data with the minimum value and the maximum value of each dimension of the search range”. The processing order of the “processing for obtaining all areas overlapping with the search range” is O (2n × a (n)) (where a is whether or not one area overlaps the search range). The processing order of “processing for comparing each dimension value of vector data with the minimum value and maximum value of each dimension of the search range” is O (r × b (n)) (r Is the number of data to be a search result, and b is the amount of processing for comparing each dimension with each dimension of the search range. For example, when the target vector data has 32 dimensions, the processing order of the “processing for obtaining all areas overlapping the search range” is O (about 4 billion × a (n)). Even if the processing time of “a” is assumed to be 1 μsec, it takes 4000 seconds only for “the process of obtaining all the regions overlapping the search range”. As described above, in the conventional method, as the number of dimensions of the target data increases, the search processing time increases. The multidimensional vector search method is applied to a search system such as a similar image search, and the search processing time required in such a search system is "about several seconds" for about 100,000 vector data. Therefore, it is considered that it is difficult to satisfy the demand by the above-mentioned conventional method.

[0005] An object of the present invention is to improve the above-mentioned problem by performing processing required for retrieving n-dimensional vector data existing in an arbitrary n-dimensional rectangular area in both position and size even when n is several tens or more. An object of the present invention is to provide a technique for reducing the influence of the number of dimensions of vector data to be searched with respect to time.

[0006]

Means for Solving the Problems In order to improve the above problem, from among a plurality of stored multidimensional vector data,
In a multi-dimensional vector search method for searching for the multi-dimensional vector data present in an arbitrary multi-dimensional rectangular area in both the position and the size, the multi-dimensional rectangular area is input as a search condition, and the multi-dimensional vector data and the multi-dimensional rectangular data are input. A first determination process is performed on a data pair consisting of an address value calculated based on an approximate value of each dimension of vector data using the address value, and when the address value satisfies the search condition, The second determination process is performed using the multidimensional vector data of the data pair, and when the vector data satisfies the search condition, the multidimensional vector data is output as a search result. That is, in the search processing according to the present invention, "first determination processing" for narrowing down data that is a search result candidate and "final search result data for the search result candidate data" are obtained. Second determination process ". The processing order of the first determination processing is O (N × f (n)) (N is the number of stored data, f is the pre-determination processing amount for each case), and the processing order of the second determination processing is Since O (N ′ × g (n)) (N ′ is the number of data to be subjected to the main determination processing and g is the main processing amount for each case), the processing order of the entire search processing is O (N
× f (n) + N ′ × g (n)). Since the address value is calculated based on the approximate value of each dimension value of the vector data, the size of the address value becomes smaller than the size of the vector data, and as a result, f (n) << g (n). Also, in the present invention, the first
By performing the determination processing of, it can be expected that the number of data to be subjected to the second determination processing is reduced to about three times the number of data as the search result. For example, in a search in which 100 search results are output from 100,000 accumulated data, the processing order O (100,000 × f (n)) of the first determination processing is compared with the processing order of the second determination processing. Is O (300 × g
(N)). Here, based on empirical observation, the processing time of f is set to 10 as the processing time taking the file I / O into consideration.
Assuming that the processing time of μ seconds and g is 1 ms, the processing time of the first determination processing is about 1 second and the processing time of the second determination processing is about 0.3 seconds. Since this is 3 seconds, it can be said that the performance is such that it can satisfy the requirement of "about several seconds" for about 100,000 vector data. Further, since the processing time of the CPU operation is much shorter than that of the file I / O, even if the number of dimensions of the vector data to be searched increases, the processing amount of the entire search processing is hardly affected.

As described above, according to the multidimensional vector search method of the present invention, even when n is several tens or more, n-dimensional vector data existing in an arbitrary n-dimensional rectangular area in both position and size can be searched. The processing time does not increase with the number of dimensions of the vector data to be searched, and can be reduced to a time that can satisfy the requirement of "about several seconds" for about 100,000 vector data. It is possible.

[0008]

Next, an embodiment of the present invention will be described with reference to the drawings.

FIG. 1 is a diagram showing the principle of a multidimensional vector search method according to the present embodiment. In the present embodiment, as a procedure for obtaining the vector data 303 existing within the search range 404 from the ten vector data 303 distributed in the multidimensional space 403, first, a first determination process is performed to obtain the ten vector data 303. From the vector data 303, the search range 40
4 is narrowed down to three vector data 303 existing in the shaded portion of the figure, which are the vector data 303 that can exist in FIG. Next, the narrowed down three vector data 303
The second determination process is performed only for
The two vector data 303 existing in are obtained. Here, since the amount of calculation in the first determination processing is very small, the time required for the search processing can be reduced.

FIG. 2 is a diagram showing the configuration of the multidimensional vector search device of the present embodiment. As shown in FIG. 2, the multidimensional vector search device according to the present embodiment includes a CPU 100, an input device 101, an output device 102, a bus 103, a memory 200, and a secondary storage device 300. When the storage processing of the present system is executed, the storage processing program 201
And an address calculation program 203 and a B-Tree program 204 are stored in the memory 200.
It is executed at 00. The vector data 303 input from the input device 101 is transferred to the memory 200, the storage processing program 201 updates the B-Tree data 301, and stores the address value 302 and the vector data 303 in the secondary storage device 300. When executing the search processing of the present system, the search processing program 202 and the address calculation program 2
03, a B-Tree program 204, an axis address calculation program 205, and an in-range determination program 206
Are stored in the memory 200 and executed by the CPU 100. Here, the B-Tree program 204 is a program for performing B-Tree processing that is generally used. The information of the search range 404 input from the input device 101 is transferred to the memory 200, and the search processing program 202
Finds the search result vector data 303 by referring to the B-Tree data 301, the address value 302, and the vector data 303, and outputs the search result vector data 303 to the output device 102. The secondary storage device 300 has B-
Tree data 301, an address value 302, and vector data 303 are stored. The B-Tree data 301 is data created by the B-Tree program 204 and, as shown in FIG. 10, corresponds to a B-Tree node part using the address value 302 as an index key. Address values 302 and B-Tree data 30 correspond to leaf portions of the B-Tree data 301.
The vector data 303 corresponds to one real data part. Hereinafter, a set of the address value 302 and the original vector data 303 from which the address value 302 is calculated is referred to as a data pair.

The storage processing program 201, the search processing program 202, the address calculation program 203, and the B-Tree program 20 for making this system function.
4. A program including the axis address calculation program 205 and the in-range determination program 206 is recorded on a recording medium such as a CD-ROM and stored in the secondary storage medium 2,
It is assumed that the program is loaded into the memory 12 and executed. The medium on which the program is recorded may be a medium other than the CD-ROM.

Hereinafter, the accumulation processing and the retrieval processing according to the present embodiment will be described.

FIG. 3 is a flowchart of the storage processing program 201 according to the present embodiment. Hereinafter, FIGS. 11 and 12
It is explained together with. In step 31, the input device 101
The address value 302 is calculated by the address calculation program 203 from the vector data 303 input from the. As shown in FIG. 11, the vector data 303 (1
When 1) is input, the address value 302 [1.3.2] of the vector data 303 (11) is calculated. In step 32, the secondary storage device 3 of the calculated address value 302
00 is stored in the B-Tree program 204
And updates the B-Tree data 301 and stores the address value 302 and the vector data 303 in the secondary storage device 3.
00 is stored. Here, since the magnitude relationship of the address values 302 is data pair (3) <data pair (11) <data pair (4), the address value 302 of the data pair (11) is set as shown in FIG. [1.3.2] is inserted between the address value 302 [1.0.3] of the data pair (3) and the address value 302 [2.0.3] of the data pair (4). , B-Tree data 301 is updated, and the address value 302 and the vector data 303 are stored in appropriate storage locations accordingly.

FIG. 4 is a flowchart of the address calculation program 203 according to this embodiment. In step 41, an address vector is created by selecting the necessary m dimensions from the vector data 303 for which the address value 302 is to be calculated, depending on the purpose of the search. In step 42, the cube 401 composed of the address vector space is set as a processing target cube. In step 43, “1” is substituted for the variable i. i is the address value 3
This is a variable for checking whether or not processing has been performed on all digits of 02. In step 44, a basic division process is performed on the target cube, and the target cube is divided into 2m subcubes 402. In step 45,
It is determined which of the subcubes 402 the vector data 303 is included in in the subcube 402 divided in step 44. In step 46, the number of the subcube 402 including the vector data 303 is set as the i-th digit Xi of the address value 302. Here, Xi is an unsigned m-bit integer. In step 47, the sub-cube 402 including the vector data 303 is set as a processing target cube. In step 48, the address value 302
Check whether all digits of have been processed. , I <k, the process proceeds to step 49. Otherwise, the process has been performed on all the digits of the address value 302, so that [X1. X2. …. Xk] is set as the address value 302 with the k numerical value sequence, and the program ends. In step 49, the variable i is incremented by "1". Here, if the number of digits k of the address value 302 is increased,
Address value 30 performed by in-range determination program 206
In the comparison process 2, the determination accuracy is improved, and the number of times the comparison process of the vector data 303 is performed can be reduced. On the other hand, the calculation amount of the comparison process itself of the address value 302,
The data amount of the address value 302 increases. for that reason,
It is necessary to set the number of digits k of the address value 302 to an appropriate value in consideration of the distribution density of the stored vector data 303 and the like.

As shown in FIG. 8, the basic division is a process of dividing an m-dimensional cube 401 into two sides, thereby dividing the cube into 2m subcubes 402. Each subcube 402 has a serial number. wear. Here, each side is divided into two equal parts at the time of division.
May be divided in consideration of the distribution bias. Address value 3
The one-dimensional ordering rule of 02 is defined such that the larger the address value 302 of the numerical value of the upper digit is, the larger the address value is. That is, when the three-digit address value 302 is calculated from the two-dimensional address vector, the address value 302 is expressed by Equation 1
When they are arranged in ascending order as shown in FIG.
In the order indicated by the arrow.

[0016]

## EQU1 ## 0.0.0.0 <0.0.1 <.
1.0 <... <3.3.2 <3.3.3 FIG. 5 is a flowchart of the search processing program 202 of the present embodiment. Hereinafter, the flow of the process performed by the search processing program on the accumulated data when the search range 404 in FIG. 13 is input will be described with reference to FIG. In step 50, the closest point and the farthest point from the origin within the search range 404 are obtained, and the address calculation program 203
Thus, the address value 302 of each point is calculated.
Hereinafter, the address value 302 of the point closest to the origin within the search range 404 is referred to as the minimum search address value and the search range 404.
The address value 302 of the point which is farthest from the origin in the above is referred to as a search maximum address value. The minimum search address value is [2.
1.1], and the search maximum address value is calculated as [3.2.1]. In step 51, the B-Tree program 2
04, the storage locations of the search minimum address value and the search maximum address value in the secondary storage device 300 are obtained. FIG.
As shown in (4), the storage location of the search minimum address value is (4)
Between (5) and (5), the storage location of the maximum search address value is (9)
And between (10). In step 52, the axis address calculation program calculates the axis address value of the minimum search address value and the axis address value of the maximum search address value. The axis address value of the minimum search address value is
The horizontal axis [0.1.1] and the vertical axis [1.0.0] are calculated,
The axis address value of the search maximum address value is expressed on the horizontal axis [1.0.
1] and the vertical axis [1.1.0]. Step 53
Then, the data pair stored in the storage position of the minimum search address value obtained in step 51 is set as the data pair to be processed. That is, the data pair (5) is set as the data pair to be processed. In step 54, it is determined whether or not it is necessary to perform processing on a data pair stored after the data pair to be processed, that is, a data pair having an address value 302 larger than the address value 302 of the data pair to be processed. . This determination is made by comparing the storage position of the search maximum address value obtained in step 52 with the storage position of the data pair to be processed. If “storage position of search maximum address value <storage position of data to be processed”, the program does not need to be continued, and the program ends. If “storage position of search maximum address value ≧ processing target data pair storage position”, the process proceeds to step 55. That is, in FIG. 14, it is not necessary to perform the determination process on the data pair (10) stored after the data pair (9). In step 55, the axis address calculation program 205 calculates the axis address of the address value 302 of the data pair to be processed.
The axis address of the address value 302 (5) is expressed by the horizontal axis [0.
1.1] and the vertical axis [1.0.1]. In step 56, the in-range determination program 206 determines whether or not the vector data 303 of the data pair to be processed exists within the search range 404. In step 57, if it is determined that the vector data 303 of the data pair to be processed exists within the search range 404, that is, if the vector data 303 of the data pair to be processed is a search result, the process proceeds to step 58. If not, proceed to step 59. In step 58, the output device 102 uses the vector data 303 of the data pair to be processed as a search result.
Output to In step 59, the process proceeds to step 54 with the data pair stored at the next position after the data pair to be processed as the processing target. When the data pair to be processed is (5),
The data pair stored next to the data pair to be processed (6)
To

As described above, the B-Tree program 2
04, the data pair (1), the data pair (2), the data pair (3), the data pair (4), and the data pair (10) are subjected to determination processing by comparing the address value 302. Since there is no need, the calculation amount of the determination process by comparing the address values 302 can be reduced to half.

FIG. 6 is a flowchart of the axis address calculation program 205 according to the present embodiment. The j-axis address value aj is used to speed up the process of evaluating the address value 302 only for the j-th vector. In step 61, “1” is substituted for a variable i. i is a variable for checking whether a binary number has been calculated for all digits of the address value 302. In step 62, the numbers xi1, xi2,..., Xim of the i-th digit Xi of the address value 302 when the number Xi is displayed in a binary number are calculated. Where m
Is the number of dimensions of the address vector. Step 6
In 3, it is checked whether binary numbers have been calculated for all digits. If i <k is satisfied, the process proceeds to step 64; otherwise, the process proceeds to step 65 since the binary numbers have been calculated for all the digits. Here, k is the number of dimensions of the address vector. In step 64, the variable i is incremented by "1". In step 65, “1” is substituted for the variable j. j is a variable for checking whether or not the axis address values in all dimensions of the address vector have been calculated. In step 66, step 62
X1j, x2j, ..., xkj among the binary numbers calculated in
Is calculated as a j-axis address value aj. In step 67, it is checked whether or not the axis address values in all dimensions of the address vector have been calculated. If j <m is satisfied, the process proceeds to step 68. Otherwise, the axis address values in all dimensions of the address vector are calculated, and the program ends. In step 68, the variable j
Is incremented by "1".

FIG. 7 is a flowchart of the in-range determination program 206 of this embodiment. Hereinafter, the flow of the process performed by the in-range determination program 206 on the accumulated data in FIG. 1 will be described with reference to FIGS. The processing performed by the in-range determination program 206 can be divided into “determination processing by comparing address values 302” and “determination processing by comparing each dimension value of vector data 303”. Steps 71 to 74 are determination processing by comparing the address values 302, and steps 75 to 78 are determination processing by comparing each dimension value of the vector data 303. In step 71, “1” is substituted for a variable i. “i” is a variable for checking whether or not the determination processing has been performed on all the axis address values of the address value 302. In step 72, it is determined whether the address value 302 of the data pair to be processed is a value from the minimum search address value to the maximum search address value. Where amin
i indicates the i-axis address value of the minimum search address value, ai indicates the i-axis address value of the address value 302 of the data pair to be processed, and amaxi indicates the i-axis address value of the maximum search address value. If it satisfies the following condition, the process proceeds to step 73. If not, the program ends, assuming that the vector data 303 of the data pair to be processed does not exist inside the search range 404. In step 73, it is checked whether or not the determination processing has been performed for all axis address values. Here, m is the number of dimensions of the address vector. If i <m is satisfied, the process proceeds to step 74. If not, the determination process is performed on all axis address values, so that the vector data 303 of the data pair to be processed exists inside the search range 404. If so, the process proceeds to step 75. In step 74, the variable i is incremented by "1". FIG. 15 shows an address value 3
02 shows the details of the determination process based on the comparison of No. 02. (5)
(5), (7), and (9) are determined by comparing the address value 302 with the data of (9) to (9) and determining that the data can exist within the search range 404. In step 75, “1” is substituted for the variable j. j is
This is a variable for checking whether or not the determination processing has been performed for all dimensions of the vector data 303. Step 7
In S6, it is determined whether or not the dimension value of the vector data 303 of the data pair to be processed is a value from the maximum value to the minimum value of the dimension in the search range 404. Here, vminj is the j-dimensional minimum value of the search range 404, vj is the j-dimensional value of the vector data 303 of the data pair to be processed, vmaxj
Indicates the maximum value of the search range 404 in the j-dimension. vmi
If nj ≦ vj ≦ vmaxj is satisfied, step 77
Proceed to. If not, the program ends, assuming that the vector data 303 of the data pair to be processed does not exist inside the search range 404. In step 77, it is checked whether or not the determination processing has been performed for all dimensions of the vector data 303. Here, n is the number of dimensions of the vector data 303. If j <n is satisfied, the process proceeds to step 78. Otherwise, the determination process is performed on all dimensions of the vector data 303. Terminates the program as it exists. In step 78, the variable j is incremented by "1". FIG.
6 shows the details of the determination process based on the comparison of each dimension value of the vector data 303. A determination process is performed on the data of (5), (7), and (9) by comparing the address value 302, and data determined to be within the search range 404 are (5) and (9). The two pieces of data are the search results.

In the present embodiment, 590-dimensional image feature values obtained by analyzing luminance differential information and color information from one image are used as vector data 303. And
A 32-dimensional address vector was created from the 590-dimensional vector data 303, and a 6-digit address value 302 was calculated. Therefore, 590-dimensional vector data 303
Address value 302 rather than calculating address value 302 from
02 becomes 32/590, and the amount of calculation required for search processing can be reduced. The size of the axis address value is 6 bits, whereas the size of each dimension value of the vector data 303 is 4 bytes = 32 bits (real value). Here, 100,000 vector data 30
By using the B-Tree program 204 for the set 3, the vector data 303 to be subjected to the determination process by comparing the address values 302 is narrowed down to の of the number of stored data, and the determination is made by comparing the address values 302. Consider a search process in which 300 cases are determined to be inside the search range 404 in the process, and 100 cases are determined to be inside the search range 404 in the determination process by comparing each dimension value of the vector data 303. In this case, the determination process by comparing each dimension value of the vector data 303 for 100,000 cases is performed by comparing the address value 302 for 50,000 cases and the determination process by comparing each dimension value of the 300 vector data 303. It can be said that it was substituted. Here, the processing amount using the B-Tree program 204 is assumed to be negligibly small as compared with other processing. The determination process by comparing each dimension value of the vector data 303 for one case is a 32-bit comparison process in each of the 590 dimensions of the vector data 303,
The determination process based on the comparison of the address value 302 for one case is a 6-bit comparison process in each of the 32 dimensions of the address vector, so the number of comparison processes per case is 32/590, and the comparison process target data is one time. The size becomes 6/32. Therefore, assuming that the determination process by comparing each dimension value of the vector data 303 for one case is 1u, the calculation amount required for the search process is reduced from 100,000 u to about 500
u + 300u = about 800u can be reduced.

As described above, according to the multidimensional vector search method of the present embodiment, n-dimensional vector data existing in an arbitrary n-dimensional rectangular area in both the position and the size is searched in the n-dimensional vector space. Can be reduced to about 1/100, and the processing amount does not increase according to the number of dimensions of the vector data to be searched. Therefore, the present invention can be applied even when n is several tens or more. It is.

[0022]

According to the present invention, in the processing for searching for n-dimensional vector data present in an arbitrary n-dimensional rectangular area in both position and size, "first determination processing" and "second determination processing" And the processing time required to search for n-dimensional vector data existing in an arbitrary n-dimensional rectangular area in both position and size even when n is several tens or more, the search target vector It is possible to reduce the influence of the number of dimensions of data.

[Brief description of the drawings]

FIG. 1 is a diagram illustrating the principle of a multidimensional vector search method according to an embodiment.

FIG. 2 is a diagram illustrating a configuration of a multidimensional vector search device according to the present embodiment.

FIG. 3 is a flowchart of an accumulation processing program 201 according to the embodiment.

FIG. 4 is a flowchart of an address calculation program 203 according to the embodiment.

FIG. 5 is a flowchart of a search processing program 202 according to the embodiment.

FIG. 6 is an axis address calculation program 205 according to the embodiment.
It is a flowchart of FIG.

FIG. 7 is a flowchart of an in-range determination program 206 of the embodiment.

FIG. 8 is a diagram illustrating basic division performed by an address calculation program 203 according to the embodiment.

FIG. 9 is a diagram illustrating a one-dimensional ordering rule of address values 302 according to the present embodiment.

FIG. 10 is a diagram illustrating details of a secondary storage device 300 according to the present embodiment.

FIG. 11 is a diagram showing a multidimensional vector space 403 for explaining the accumulation processing of the present embodiment.

FIG. 12 is a diagram showing details of a secondary storage device 300 for explaining an accumulation process of the present embodiment.

FIG. 13 is a diagram illustrating a multidimensional vector space 403 for describing search processing according to the present embodiment.

FIG. 14 is a diagram illustrating details of a secondary storage device 300 for describing a search process according to the present embodiment.

FIG. 15 is a diagram illustrating detailed contents of a determination process by comparing the address values 302 according to the present embodiment.

FIG. 16 is a diagram illustrating detailed contents of a determination process based on a comparison of each dimension value of the vector data 303 according to the present embodiment.

[Explanation of symbols]

 Reference Signs List 101 CPU 101 input device 102 output device 103 bus 200 memory 201 storage processing program 202 search processing program 203 address calculation program 204 B-Tree program 205 axis address calculation program 206 in-range determination program 300 secondary storage device 301 B-Tree data 302 Address value 303 Vector data 401 Cube 402 Subcube 403 Multidimensional vector space

Claims (6)

[Claims] 1. A multi-dimensional vector search method for searching, among a plurality of stored multi-dimensional vector data, the multi-dimensional vector data present in an arbitrary multi-dimensional rectangular area in both position and size. A rectangular area is input as a search condition, and for a data pair consisting of the multidimensional vector data and an address value calculated based on the approximate value of each dimension of the multidimensional vector data, a data pair using the address value is used. 1 is performed, and if the address value satisfies the search condition, a second determination process is performed using the multidimensional vector data of the data pair, and if the vector data satisfies the search condition, A multidimensional vector search method, comprising outputting the multidimensional vector data as a search result. 2. The multi-dimensional vector search method according to claim 1, wherein, when storing the multi-dimensional vector data, the address value calculated based on an approximate value of each dimension of the multi-dimensional vector data and A multi-dimensional vector search method characterized by storing multi-dimensional vector data separately in association with each other. 3. The multi-dimensional vector search method according to claim 1, wherein, when calculating the address value, an address vector is created by selecting an arbitrary dimension of the multi-dimensional vector data. A multidimensional vector search method, characterized in that a value calculated based on an approximate value of each dimension of a vector is used as an address value. 4. A multi-dimensional vector search method according to claim 1, wherein a rule for assigning a one-dimensional order to said address value is set, and a B-Tree using said address value as an index key is created to create said data pair. Is stored, and when the first determination process is performed, the minimum address value of the search range and the storage position of the maximum address value of the search range are determined using the B-Tree, and the minimum address value of the search range is determined. Wherein the first determination process is not performed on the data pair having an address value smaller than and the data pair having an address value larger than the maximum address value in the search range. retrieval method. 5. A multidimensional vector retrieval method for retrieving, from a plurality of accumulated multidimensional vector data, said multidimensional vector data present in an arbitrary multidimensional rectangular area in both position and size. A search condition input unit for inputting a rectangular area as a search condition, and a data pair consisting of the multidimensional vector data and an address value calculated based on an approximate value of each dimension of the multidimensional vector data, the corresponding address 1st performing the 1st judgment processing using the value
A second determination processing unit that performs a second determination process using the multidimensional vector data of the data pair when the address value satisfies the search condition; A search result output unit that outputs the multidimensional vector data as a search result when a condition is satisfied. 6. A multidimensional vector search method for searching, among a plurality of accumulated multidimensional vector data, the multidimensional vector data present in an arbitrary multidimensional rectangular area in both position and size. A search condition input unit for inputting a rectangular area as a search condition, and a data pair consisting of the multidimensional vector data and an address value calculated based on an approximate value of each dimension of the multidimensional vector data, the corresponding address 1st performing the 1st judgment processing using the value
A second determination processing unit that performs a second determination process using the multidimensional vector data of the data pair when the address value satisfies the search condition; A medium in which a program for causing a computer to function as a search result output unit that outputs the multidimensional vector data as a search result when a condition is satisfied is recorded.