JP3708809B2 - Recording medium on which multidimensional spatial data structure is recorded, multidimensional spatial data update method, multidimensional spatial data search method, and recording medium on which program for executing the method is recorded - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a search technique for multidimensional spatial data stored in a database, and in particular, a recording medium recording a multidimensional spatial data structure, a multidimensional spatial data update method, a multidimensional spatial data search method, and a program for executing the method The present invention relates to a recording medium that records
[0002]
[Prior art]
Among conventional spatial access methods, R-tree and its derivatives are useful methods for searching a high-dimensional space (hereinafter referred to as R-tree family). The R-tree family includes R-tree, R^*There are -tree, Hilbert R-tree, SS-tree, SR-tree, etc., and the data object is a minimum enclosing area, that is, a minimum enclosing rectangle (MBR) or a minimum enclosing sphere (MBS). In addition, it has a hierarchical structure.
[0003]
SR-tree has a tree structure using MBR and MBS as shown in FIG. In the search, the distance between the inquiry point and the MBR and the distance between the inquiry point and the MBS are calculated, and the longer distance is set as the distance between the entry and the inquiry point. By using MBR and MBS for search, the pruning process is made more efficient.
[0004]
There is a VR-tree (Japanese Patent Application No. 2000-8056) as a method using the encoding of MBR position coordinates. The MBR is approximated by a virtual bounding rectangle (VBR), the data object is approximated by a relative cell, and the VBR and the relative cell are expressed by a subspace code.
[0005]
FIG. 8 is a simple illustration of a method for calculating a subspace code. In the i-dimensional coordinates in the n-dimensional space, the surrounding rectangle A occupies (3, 19) as a coordinate value, and the surrounding rectangle B similarly occupies (6, 8) as a coordinate value. At this time, the i-dimensional coordinate range of A is set to 2 using a code of length l.^lIs equally divided into partial sections, the starting point of B can be approximately represented by the partial section in which it exists. The same applies to the end point of B. For example, if a region A is divided into eight partial sections using a code having a length of 3, B occupies the second and third partial sections from the beginning counted from the start point of A. Therefore, the region B in i-dimensional coordinates can be expressed approximately as an 8-ary code (1, 2) or a binary code (001, 010). In this case, since the code length required for the binary code is 3 for each of the start point and the end point of B, there are a total of 6 per dimension.
[0006]
In VR-tree, two types of tree structure indexes are constructed. It consists of a real part which is a tree structure using absolute position representation and a virtual part which is a tree structure using relative position expression based on subspace codes. FIG. 9 shows an example of the index structure. In FIG. 9A, the entire space is R. MBR M1 and M2 are included in R, and V1 and V2 approximate M1 and M2 based on R, respectively. V1 and V2 are V1 of M1 in R and VBR of M2 in R, respectively. Similarly, in FIG. 9B, V3 and V4 are V3 of M3 in V1 and VBR of M4 in V1, respectively. Further, as shown in FIG. 9C, the data objects P1 and P2 are included in M3. C1 and C2 approximate P1 and P2 based on V3, respectively. C1 and C2 are P1 relative cells in V3 and P2 relative cells in V3, respectively. The MBR, VBR, data object, and relative cell shown in FIGS. 9A, 9B, and 9C are configured as shown in FIG. MBR and data object are stored in the real part node, and VBR or relative space code of the relative cell is stored in the virtual part node. The virtual part node and the real part node maintain a one-to-one relationship.
[0007]
[Problems to be solved by the invention]
The conventional VR-tree described above has the following problems.
[0008]
When a data object is newly inserted into the index or deleted, an index update process occurs. In VR-tree, it is necessary to update not only the real part but also the virtual part. Therefore, more update costs are generated than in the conventional method. In practical use, there are cases where updates frequently occur, so it is desirable to reduce the update cost.
[0009]
Specifically, the outline of the update process in VR-tree and the reason why a lot of update costs occur will be described. While the MBR correction process in the real part propagates upward, the modification of the subspace code in the virtual part propagates downward. When the MBR or data object is updated in the real part, the corresponding VBR or relative cell code is also updated. Furthermore, it is necessary to access all nodes located under the updated code entry as a vertex and check the subspace code held by the node. And update if necessary. The procedure is as follows.
[0010]
(1) In order to update the code of the VBR corresponding to the updated MBR or the code of the relative cell corresponding to the updated data object, the VBR is calculated from the root node toward the lower node in the virtual part.
[0011]
(2) The code corresponding to the updated MBR or data object is calculated and updated.
[0012]
(3) Access all the nodes in the subtree under the updated entry, check the codes stored in the nodes, and update if necessary.
[0013]
As described in (3), in VR-tree, even if a node that does not require modification of the subspace code is reached, the process does not end, but the process is continuously propagated to the lower nodes, and is transmitted to all nodes in the subtree. to access. This is because the subspace code uses a relative position, so the VBR contains some error, and even if there is no need to update at the current node, it may need to be modified at the lower node. Because there is. The reason for this will be described with reference to FIG. In FIG. 9, rectangles A, B, and C are respectively defined as the entire space, the child node MBR, and the grandchild node MBR. Also, V_bIs VBR of B in radix 4 A, and V_cRadix 4_bIt is assumed that V BR of C in FIG. Assume that only the rectangle A is expanded in the x-axis direction as shown in FIGS. 9A to 9B by inserting the data object. At this time, there is no change in the size of B._bExpands. And for C with no variation in size, V_bVBR V of C due to size variation of_cShrinks. In this case, V_bThere is no need to modify the subspace code of_cThis subspace code needs to be modified.
[0014]
VR-tree requires additional update costs as described above. If the MBR at the upper level of the tree structure is changed, a particularly high cost is required.
[0015]
The present invention has been made in view of the above, and an object of the present invention is to provide a recording medium on which a multidimensional spatial data structure that can search for an object in a short time and has high search performance is recorded, and a multidimensional spatial data update method Another object of the present invention is to provide a multidimensional spatial data search method and a recording medium on which a program for executing the method is recorded.
[0016]
[Means for Solving the Problems]
  In order to achieve the above object, a recording medium recording the multidimensional spatial data structure of the present invention according to claim 1 is a multidimensional space for organizing multimedia data stored in a database.Recording medium on which data structure is recordedBecauseThe multidimensional spatial data structure includes a root node that stores a virtual enclosing rectangle and a centroid, andLink as a child node to the root nodeStores the virtual enclosing rectangle that approximates the minimum enclosing rectangle, the minimum enclosing rectangle, and the centroid of the underlying data objectAn intermediate node,Linked to this intermediate node as a child node, three types of leaf nodes that store relative cells approximating the data object and a minimum enclosing rectangleAn index node with nodes and,Link to the leaf nodes in a one-to-one relationshipStore data objectsConsists of data nodes andBeThis is the gist.
[0017]
  According to the first aspect of the present invention, a data object is stored in a data node, a relative cell and a minimum enclosing rectangle that approximate the data object are stored in a leaf node, and a virtual enclosing that approximates a minimum enclosing rectangle in an intermediate node Since the multidimensional spatial data structure configured to store the rectangle, the minimum enclosing rectangle, and the center of gravity of the subordinate data object and store the virtual enclosing rectangle and the center of gravity in the root node is recorded on the recording medium, the recording medium make use of,Multidimensional spatial data structure that organizes multimedia data stored in a databaseCan improve the distribution of
[0018]
  Further, the multidimensional spatial data update method of the present invention described in claim 2 is the method described in claim 1.Recorded on a recording mediumMultidimensional spatial data structureA multidimensional spatial data update method in whichThe data object stored in the data nodeOf the data nodeCalculate the subspace code of the relative cell from the minimum enclosing rectangle stored in the parent node, and the minimum enclosing rectangle stored in the index node other than the root node and the minimum enclosing rectangle stored in the parent node of the index node The main point is to calculate the subspace code of the virtual enclosing rectangle from the above.
[0019]
  In the present invention, the subspace code of the relative cell is calculated from the data object stored in the data node and the minimum enclosing rectangle stored in the parent node, and the index node other than the root node is calculated. Calculate the subspace code of the virtual enclosing rectangle from the stored minimum enclosing rectangle and the minimum enclosing rectangle stored in the parent nodeTo update the index associated with data object insertion and reduction at low cost.
[0020]
  Furthermore, the multidimensional spatial data search method of the present invention as set forth in claim 3 is the method according to claim 1.Recorded on a recording mediumMultidimensional spatial data structureA multidimensional spatial data search method in whichIn the root node and the intermediate node, the virtual enclosing rectangle is calculated from the minimum enclosing rectangle and the subspace code, and in the leaf node, the relative cell is calculated from the minimum enclosing rectangle and the subspace code, and the data object is obtained by accessing the data node. The gist.
[0021]
  In the present invention described in claim 3, the root node and the intermediate node calculate a virtual enclosing rectangle from the minimum enclosing rectangle and the subspace code, and the leaf node calculates a relative cell from the minimum enclosing rectangle and the subspace code, Get data object in data node accessThis makes it possible to detect an object in a short time and realize high search performance.
[0022]
  A recording medium on which the multidimensional spatial data update program of the present invention described in claim 4 is recorded comprises:Recorded on a recording mediumMultidimensional spatial data structureA recording medium recording a program for causing a computer to execute the multidimensional spatial data update method in the computer,The data object stored in the data nodeOf the data nodeCalculate the subspace code of the relative cell from the minimum enclosing rectangle stored in the parent node, and the minimum enclosing rectangle stored in the index node other than the root nodeOf the index nodeThe gist is to calculate the subspace code of the virtual enclosing rectangle from the minimum enclosing rectangle stored in the parent node.
[0023]
  In the present invention described in claim 4, the subspace code of the relative cell is calculated from the data object stored in the data node and the minimum enclosing rectangle stored in the parent node, and the index node other than the root node is calculated. Since a data update program for multidimensional spatial data for calculating a subspace code of a virtual enclosing rectangle from the stored minimum enclosing rectangle and the minimum enclosing rectangle stored in the parent node is recorded on the recording medium, the recording medium Using,Multi-dimensional spatial data update program that performs index update at low cost due to insertion and reduction of data objectsCan improve the distribution of
[0024]
  A recording medium on which the multidimensional spatial data search program of the present invention is recorded.Recorded on a recording mediumMultidimensional spatial data structureA recording medium on which a program for causing a computer to execute the multidimensional spatial data search method is recorded,In the root node and the intermediate node, the virtual enclosing rectangle is calculated from the minimum enclosing rectangle and the subspace code, and in the leaf node, the relative cell is calculated from the minimum enclosing rectangle and the subspace code, and the data object is obtained by accessing the data node. The gist.
[0025]
  In the present invention of claim 5, the root and intermediate nodes calculate a virtual enclosing rectangle from the minimum enclosing rectangle and the subspace code, and the leaf node calculates a relative cell from the minimum enclosing rectangle and the subspace code, Since a multidimensional spatial data search program for acquiring a data object in data node access is recorded on a recording medium, the recording medium is used,Multidimensional spatial data search program that detects objects in a short time and realizes high search performanceCan improve the distribution of
[0026]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the present invention, in view of the problems of the VR-tree described above, (1) the update process calculates a subspace code from the parent node's MBR instead of the parent node's VBR, and (2) the search process has few An A-tree (Approximation Tree) improved from VR-tree, which is a new indexing technique having a data structure and algorithm having characteristics of acquiring MBR and subspace code data by page access, has been proposed.
[0027]
In this A-tree, the VBR and relative cell subspace code are calculated relative to the MBR of the upper node. Also, the subspace code and the MBR of the parent node are stored in the same node.
[0028]
FIG. 1 is a diagram showing an example of the structure of this A-tree. In FIG. 1A, the entire space is R. MBR M1 and M2 are included in R, and MBR M3 and M4 are included in M1. Further, the data objects P1 and P2 are included in M3. In this structure, V1, V2, V3, and V4 are M1's VBR in R, M2's VBR in R, M3's VBR in M1, and M4's VBR in M1. C1 and C2 are P1 relative cells in M3 and P2 relative cells in M3, respectively. The MBR, VBR, data object, and relative cell are stored as shown in FIG. 1B, and a tree structure is created.
[0029]
As shown in FIG. 1B, the A-tree nodes can be classified into an index node and a data node. The index nodes other than the root node have one MBR (M) and a VBR (or relative cell) subspace code approximating the MBR (or data object) related to the child node. M is an MBR including the MBR (or data object) stored in the child node. MBR is not stored in the root node. The root node can calculate the VBR of the child node from the entire space. When building an A-tree, the sign of the child's VBR (or relative cell) stored in a node is calculated from the parent's MBR stored in the same node as the child MBR (or data object). . At the root node, the sign of the child VBR is calculated from the child MBR and the entire space.
[0030]
In A-tree, index nodes are classified into leaf nodes, intermediate nodes, and root nodes.
[0031]
1. Data node
Data node is entry, (P₁, O₁), (P₂, O₂), ..., (P_m, O_m) Is included. Where P_i(I = 1, 2,..., M) is a space vector of the data object, and o_i(I = 1, 2,..., M) is a pointer to the object description record. The number of entries m is limited by the maximum value and the minimum value.
[0032]
2. Leaf node
Leaf nodes have a one-to-one relationship with data nodes, and data nodes N ((P₁, O₁), (P₂, O₂), ..., (P_m, O_m)) And the corresponding leaf node i) P₁, P₂, ..., P_mM, ii) pointer N, iii) list of entries sc (V₁), Sc (V₂), ..., sc (V_m). Where V_iIs P in M_i(I = 1, 2,..., M) relative cell, SC (V_i) Is V_iIs a subspace code.
[0033]
3. Intermediate node
The index node excluding the root node and leaf node is an intermediate node, and i) MBR M including the MBR of the child node, ii) entry (ptr, sc (V), w, P)_centroid) List. Here, ptr is a pointer to the child node C, V is a VBR approximating the MBR stored in C, w is the total number of data objects in the subtree having C as a vertex, P_centroidIs the centroid of all data objects in the subtree.
[0034]
w and P_centroidIs used when inserting and deleting data objects. Since only M, ptr, and sc (V) are used in the search algorithm, ptr and sc (V) data are stored in the same page in the implementation.
[0035]
4). Root node
The root node is an entry (ptr, sc (V), w, P_centroid) List.
[0036]
Next, the above-described A-tree update algorithm will be described. The update algorithm for A-tree is based on SR-tree. Starting from the insertion or deletion of the data object, the MBR and centroid adjustment processing at the intermediate node is propagated to the upper node. The difference between A-tree and SR-tree is that the A-tree algorithm requires calculation of VBR and relative cells. Specifically, the A-tree structure is updated as follows.
[0037]
(1) Let N be a data node and M be an MBR of N. At that time, M is stored in the parent node of N. When a data object is inserted or deleted in N, M is adjusted, and the centroids of all data objects stored in N are calculated and updated.
[0038]
(2) If M does not change, the sign of the relative cell approximating the insertion object is calculated from M and updated. Otherwise, update all codes stored in the N parent nodes.
[0039]
(3) Let N be one of the index nodes excluding the root node, and let M be the MBR of N. At that time, M is stored in the parent node of N. When insertion or deletion of a data object occurs in a subtree having N as a vertex, all barycentric points accumulated in N parent nodes are updated. Further, if the insertion or deletion changes the child MBR, M is adjusted.
[0040]
(4) If M does not change, the code of VBR that approximates the updated MBR is calculated from M. And update. Otherwise, it updates the sign of all VBRs stored in the N parent node.
[0041]
Next, a neighborhood search algorithm will be described using the above-described A-tree with reference to the flowchart shown in FIG. This search algorithm is based on VR-tree. The difference is that VR-tree calculates the VBR and relative cell coordinate values from the parent node's VBR, whereas A-tree calculates it from the parent node's MBR.
[0042]
In the search algorithm shown in FIG. 2, a tree structure is searched with width priority. Based on the MBR of the current node and the subspace code of the child node, the VBR of the child node is calculated at the root node and the intermediate node, and the relative cell of the child node is calculated at the leaf node. When the data node is reached, the position information of the object is collected and set as a candidate object.
[0043]
In FIG. 2, first, a root node is set at the head of the queue (step S11), and the processing of loop 1 of steps S13 to S39 is repeatedly executed until the queue becomes empty. In this loop 1, first, the pointer to the node is set to N, the head data of the queue is designated by this pointer N, taken out from the queue as the data related to the node closest to the inquiry point, and in the node indicated by the pointer N The entry number is set to e (step S15). Then, it is determined whether the node indicated by the pointer N is a data node or an index node (step S17).
[0044]
If the node pointed to by N is a data node, the data node occupies multiple pages, so only the entry that is pointed to, ie the page for the data object, is accessed. Specifically, the e-th entry stored in the node pointed to by N is set as E, the distance between the data object that is the entry E and the inquiry point is calculated, and the calculated value is D (step S19).
[0045]
This calculated distance D is compared with the k vicinity distance of the first list, and it is checked whether or not the distance D is equal to or less than the k vicinity distance of the first list (step S21). If the distance D is small, the calculated distance D and the data object E are stored in the first list as the nearest candidate and sorted (step S23). Then, the queue is filtered by the first list (step S25). In this filtering, data that does not need to be accessed in the queue is deleted using candidate data objects stored in the first list.
[0046]
Then, the head data of the queue is deleted, the data in the queue is prepended (step S27), and the above processing is repeated as loop 1 until the queue becomes empty.
[0047]
If the node indicated by N is an index node in the determination at step S17, the process proceeds to step S31, where the number of entries stored in the node indicated by N is set as ent, Set N virtual bounding rectangle MBR to R. Then, it is determined whether or not the node indicated by N is a leaf node (step S33).
[0048]
If the node indicated by N is not a leaf node but an intermediate node or root node, the process A shown in FIG. 3 is executed (step S35), and if it is a leaf node, the process B shown in FIG. Is executed (step S37).
[0049]
In process A shown in FIG. 3, first, parameter i is set to 0 (step S51), and the process shown in loop 2 from step S53 to S65 is repeated until parameter i is equal to or greater than the number of entries ent. . In loop 2, first, the parameter i is incremented by +1 (step S55), and the VBR (virtual enclosing rectangle) V of the child node is obtained from the subspace code stored as the i-th entry in the node N and the parent node rectangle R. Calculate (step S57). Then, the minimum distance D between the calculated V and the inquiry point is calculated (step S59), the distance D is compared with the k neighborhood distances of the first and second lists, and the distance D is the k neighborhood distance of the first list. It is determined whether or not it is less than or equal to the k vicinity distance of the second list (step S61).
[0050]
As a result of this determination, the child pointer and distance stored in the i-th entry of the node N only when the distance D is less than or equal to the k neighborhood distance of the first list and less than or equal to the k neighborhood distance of the second list D is stored in the queue (step S63). The above loop 2 processing is performed on all entries stored in the node, and when loop 2 ends, the head data in the queue is deleted, the data is left-justified, and the data in the queue is sorted in ascending order of distance. Sorting is performed (step S67), the process returns from step A in FIG. 3 to step S39 shown in FIG. 2, and loop 1 is repeated until the queue becomes empty.
[0051]
Next, in the process B shown in FIG. 4, first, the parameter i is set to 0 (step S81), and the process shown in the loop 3 from the steps S83 to S101 that follow is performed until the parameter i reaches the entry number ent or more. Run repeatedly. In loop 3, first, the parameter i is incremented by +1 (step S85), and the relative cell C of the child node is calculated from the subspace code stored as the i-th entry in the node N and the rectangle R of the current node (step S85). S87). Then, the minimum distance D between the calculated relative cell C and the inquiry point is calculated (step S89), and this distance D is compared with the k-neighbor distances in the first and second lists, and the distance D is k in the first list. It is determined whether or not the distance is less than or equal to the neighborhood distance and less than or equal to the distance k near the second list (step S90).
[0052]
If the result of this determination is that the distance D is less than or equal to the k neighborhood distance of the first list and less than or equal to the k neighborhood distance of the second list, the child pointer stored in the i th entry of the node N, entry number i The distance D is stored in the queue and sorted (step S91).
[0053]
Next, Dmax which is the maximum distance between the calculated relative cell C and the inquiry point is calculated (step S93). It is determined whether or not the calculated maximum distance Dmax is less than the k vicinity distance of the second list (step S95). If the maximum distance Dmax is less than the k vicinity distance of the second list, the maximum distance Dmax is determined. Are stored in the second list and sorted (step S97). Then, queue filtering is performed using the second list (step S99). In this filtering, the candidate objects stored in the second list are used to delete data that does not need to be accessed in the queue, thereby reducing the amount of data in the queue. The above loop 3 processing is performed for all entries stored in the node, and when loop 2 ends, the head data in the queue is deleted, the data is left-justified, and the data in the queue is sorted in ascending order of distance. Sorting is performed (step S67), the process returns from step B in FIG. 4 to step S39 shown in FIG. 2, and loop 1 is repeated until the queue becomes empty.
[0054]
In this way, the process of loop 1 is repeatedly executed until the queue becomes empty, and when the process of loop 1 is completed, the candidate objects stored in the first list are output as final neighborhood objects (step S41). .
[0055]
The multidimensional spatial data update process and multidimensional spatial data search process described above are performed by a computer comprising a CPU 1, a memory 3, an I / O 5, an input means 7, a display means 9, and an external storage means 11, as shown in FIG. The CPU 3 executes the multidimensional spatial data update program and the multidimensional spatial data search program stored in the memory 3 or the external storage unit 11.
[0056]
More specifically, the computer shown in FIG. 5A includes an index update processing unit as shown in FIG. 5B configured as a software function by the multidimensional spatial data update program, the multidimensional spatial data search program, and the CPU 1. 21 and the search processing unit 23, the index update processing unit 21 accesses the index storage unit 25 configured in the memory 3, updates the index stored in the index storage unit 25, and performs search processing. The unit 23 performs a search while accessing the index storage unit 25.
[0057]
Here, a search algorithm will be described as an example with reference to FIG. V1 and V2 are obtained from the rectangular position of R and the subspace codes of V1 and V2, and the distance between VBR and the inquiry point is calculated and stored in the queue. When V1 is not a target for pruning, V3 and V4 are similarly obtained from the position of M1 and the subspace codes of V3 and V4, and the distance between VBR and the inquiry point is calculated. The distance obtained by calculation is stored in the queue. When V3 is not a target for pruning, C1 and C2 are calculated based on the position of M3 and the subspace codes of C1 and C2. The distance between the relative cell obtained by calculation and the inquiry point is stored in the queue. The maximum distance between the inquiry point and C1, and the maximum distance between the inquiry point and C2 are stored in the second list, and the queue is filtered. If C1 is not a pruning target, P1 is accessed.
[0058]
Next, the experimental result of investigating the search time by the multidimensional spatial data search method of the present invention will be described with reference to FIG. The SR-tree and VA-File of the existing method were used as comparison targets for the present invention in this experiment. This SR-tree, VA-File is an excellent technique in high-dimensional space search. As experimental conditions, the object data is point data based on the hue histogram of the image, the data set size is 100,000, and the number of searches is 20. The number of spatial dimensions was changed from 4 dimensions to 64 dimensions. The disk page was based on 8KB (8192 bytes).
[0059]
As can be seen from FIG. 6A, A-tree is superior to VA-File and SR-tree in terms of search cost in all dimensions. Specifically, in the case of 100,000 cases and 64 dimensions, a disk access reduction of 77.7% was achieved for VA-File, and 77.3% of disk access was achieved for SR-tree. Reduction has been achieved. VR-tree shows equivalent performance.
[0060]
Further, as can be seen from FIG. 6B, the index update cost is lower than that of VR-tree. Specifically, in the case of 100,000 cases and 64 dimensions, the disk access reduction of 68.2% is achieved with respect to the VR-tree.
[0061]
The processing procedures of the space search method and index update method of the above embodiment are recorded as a program on a recording medium, and the recording medium is incorporated into a computer system, and the program recorded on the recording medium is downloaded to the computer system. Of course, by installing and operating the computer system with the program, it can function as a system for performing the space search method and the index update method. It can be raised.
[0062]
【The invention's effect】
As described above, according to the present invention, an object can be detected in a short time in a high-dimensional space search, and high search performance can be realized. In addition, it is possible to update the index associated with insertion and reduction of data objects at a low cost. Furthermore, the present invention can be applied to a wide range of information retrieval for images, videos, sounds, documents, etc. that can be expressed as multidimensional spatial data.
[Brief description of the drawings]
FIG. 1 is a diagram illustrating an example of a structure of an A-tree that is a multidimensional spatial data structure according to an embodiment of the present invention.
FIG. 2 is a flowchart illustrating an operation of a multidimensional spatial data search method according to another embodiment of the present invention.
FIG. 3 is a flowchart showing a procedure of processing A in step S35 in the flowchart shown in FIG. 2;
4 is a flowchart showing a procedure of a process B in step S37 in the flowchart shown in FIG.
FIG. 5 is a diagram showing a system configuration for performing multidimensional spatial data update processing and multidimensional spatial data search processing according to an embodiment of the present invention.
FIG. 6 is a graph showing an experimental result obtained by investigating a search time according to the multidimensional spatial data search method of the present invention in comparison with the prior art.
FIG. 7 is a diagram showing the structure of SR-tree in the prior art.
FIG. 8 is a diagram illustrating a subspace code calculation method in the prior art.
FIG. 9 is a diagram showing a VR-tree structure in the prior art.
FIG. 10 is a diagram for explaining VR-tree update processing in the prior art;
[Explanation of symbols]
1 CPU
3 memory
11 External storage means
21 Index update processing section
23 Search processing section
25 Search storage unit
C1, C2 relative cell
M1, M2, M3, M4 MBR (minimum enclosure rectangle)
P1, P2 data objects
V1, V2, V3, V4 VBR (virtual enclosing rectangle)

Claims (5)

A recording medium recording a multidimensional spatial data structure that organizes multimedia data stored in a database,
The multidimensional spatial data structure is
The root node that stores the virtual enclosing rectangle and the center of gravity, and the virtual enclosing rectangle that approximates the minimum enclosing rectangle, the minimum enclosing rectangle, and the center of gravity of the subordinate data object are stored. An index node having three types of nodes : an intermediate node and a leaf node linked to the intermediate node as a child node and storing a relative cell approximating the data object and a minimum enclosing rectangle ;
The link in a one-to-one relationship to the leaf node, recording medium recording a multi-dimensional space data structure, characterized in that it is composed of a data node which stores the data objects. A multi-dimensional spatial data update method definitive multidimensional spatial data structure recorded on a recording medium according to claim 1,
Calculate the subspace code of the relative cell from the data object stored in the data node and the minimum enclosing rectangle stored in the parent node of the data node, and the minimum enclosing rectangle stored in the index node other than the root node A multidimensional spatial data update method, comprising: calculating a subspace code of a virtual enclosing rectangle from a minimum enclosing rectangle stored in a parent node of the index node . A multi-dimensional space data search method definitive multidimensional spatial data structure recorded on a recording medium according to claim 1,
In the root node and the intermediate node, the virtual enclosing rectangle is calculated from the minimum enclosing rectangle and the subspace code, and in the leaf node, the relative cell is calculated from the minimum enclosing rectangle and the subspace code, and the data object is obtained by accessing the data node. A characteristic multidimensional spatial data search method. A recording medium recording a program for executing a multi-dimensional space data updating method definitive multidimensional spatial data structure recorded on a recording medium according to claim 1, wherein the computer,
Calculate the subspace code of the relative cell from the data object stored in the data node and the minimum enclosing rectangle stored in the parent node of the data node, and the minimum enclosing rectangle stored in the index node other than the root node A recording medium on which a multidimensional spatial data update program is recorded, wherein a subspace code of a virtual enclosing rectangle is calculated from a minimum enclosing rectangle stored in a parent node of the index node . A recording medium recording a program for executing a multi-dimensional space data search method definitive multidimensional spatial data structure recorded on a recording medium according to claim 1, wherein the computer,
In the root node and the intermediate node, the virtual enclosing rectangle is calculated from the minimum enclosing rectangle and the subspace code, and in the leaf node, the relative cell is calculated from the minimum enclosing rectangle and the subspace code, and the data object is obtained by accessing the data node. A recording medium on which a characteristic multidimensional spatial data search program is recorded.

Images