JPH10261965A - Coding method based on structuring of multi-dimensional data - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a coding method for multi-dimensional data in which coding, processing and visualization of a required part of data are easily conducted based on structuring of the data that facilitates retrieval for position information of contents of the data. SOLUTION: A data structuring part 3 generates a space division model of a data area division result, assigns an area attribute to each node of a graph illustrating the model and an area location information coding part 8 codes the area attribute. In the case of coding area internal information of the data, an area location retrieval part 5 obtains location information of a required part of the data based on the retrieval of the graph illustrating the space division model. Then an information processing part 6 processes internal information of the area and an area inside information coding part 7 encodes the information by a method adaptive to an attribute of the area.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compression encoding method in data communication.

[0002]

2. Description of the Related Art Data having three or more dimensions (hereinafter, referred to as multidimensional data) has an enormous amount of information, and therefore requires an encoding method for transmission and storage.

Conventionally, encoding of multidimensional data has been performed using vector quantization, DCT, wavelet, fractal, etc. (P. Ning and L. Hesslink: "Fast Volume Renderin").
gof Compressed Data, "Proc.of IEEE Visualization'9
3, pp.11-18,1993, B.Yeo and B.Liu: "Volume Rendering
of DCT-Based Compressed 3D Scalar Data, "IEEETrans,
on Visualization and Computer Graphics, Vol.1, pp.2
9-43,1995, S.Muraki: "Volume data and wavelet transf
orms, "IEEE Computer Graphics & Applications, Vol.1
3, No.4, pp.50-56,1993, WOCochran, JCHart and PJ
Flynn: "Fractal Volume Compression," IEEE Trans.on V
isualization and Computer Graphics, Vol.2, No.4, pp.3
13-322, 1996).

[0004] In these methods, waveform coding is directly applied to a three-dimensional data array without structuring data using region division or the like. In other words, data contents, such as bones and organs for medical data, are extracted from the data and are not encoded in a state including their position information and the like.

However, in coding multidimensional data, it is important to structure the data. this is,
Since the contents of multidimensional data cannot be directly displayed on a two-dimensional display, some kind of visualization method is required, and in order to appropriately visualize a desired part, the structure of the data using region division or the like is required. This is because conversion and data processing are indispensable.

Hitherto, many coding methods involving data structuring have been proposed in video coding methods (for example, Japanese Patent Application Laid-Open No. H7-154699).

[0007]

However, in the case of a video or the like, since the data of each time-series image is two-dimensional, the purpose of structuring the data by region division or the like is as follows.
The main purpose is to improve the estimation accuracy of a motion vector between frames.

Since two-dimensional data can be directly displayed on a display, visualization is not necessary.
In Japanese Patent No. 154699, the relationship between visualization and coding is not considered, and it is difficult to directly extend the method proposed for video to multidimensional data.

In order to make the method suitable for multi-dimensional data, it is necessary to consider its desirable properties for encoding and visualization. One of its desirable properties is the ease of searching for location information of data contents. Multidimensional data has a very large amount of data due to its high dimension, and therefore requires a large amount of calculation in both encoding and visualization. However, it is not always necessary to encode or visualize the entire data.

[0010] Also, being able to encode and visualize only necessary parts is important for improving the speed of the system when finally presenting data. Therefore, where necessary parts are located in the entire data. There is a need for knowing whether to do so, that is, for making it easy to search for data position information.

[0011] From the above, it is desired that a coding method for multidimensional data that requires visualization has a data structure that makes it easy to search for position information of data contents.

Further, the structuring of multidimensional data is not only performed by coding, but also by using a database storing a large amount of data.
It is also useful when searching for specific data. That is, even when searching for data having a specified attribute in a specified portion, structuring of multidimensional data is desired from the ease of searching for position information of data contents.

SUMMARY OF THE INVENTION It is an object of the present invention to provide multi-dimensional data encoding based on data structuring that makes it easy to search for position information of data contents.

[0014]

In order to solve the above problems, according to the present invention, the content is extracted through the division of multi-dimensional data to be encoded, and the position information and internal information of each region are individually obtained. In the encoding method, when encoding the position information of each region, a method of generating a space division model of the region represented by the graph, assigning a region attribute to each node of the graph, and encoding the attribute is proposed. Is what you do.

When assigning a region attribute to each node of the graph, information obtained from a result of the region division and information of an object included in each region may be attached as the attribute.

Further, according to the present invention, when encoding the internal information of each area, an area search is performed based on a graph representing the space division model, and the search result is used to assign an area attribute to each area. This is to perform adaptive encoding.

[0017]

According to the present invention, since the position information of the content is stored in the coded information by the graph, the search for the position is replaced with the search for the graph, and the search for the position information of the multidimensional data is performed. It can be performed easily and at high speed, and as a result, coding and visualization of necessary parts can be easily performed.

Further, when encoding the internal information of a region, the attribute of the region to be encoded can be determined from the attributes of the nodes of the graph representing the space division model of the region. Can be performed.

Further, according to the present invention, when searching a database storing a large amount of encoded multidimensional data,
First, only the graph attached to each data is decoded, and then, by using the area search based thereon, the search can be performed by specifying the attribute of an arbitrary part of the data, and based on the result, the data can be searched. It is possible to decode only part of the internal information and present it as a search result candidate through visualization.

Further, an information processing operation can be applied to each area on the information sender side and the information receiver side by using area search based on a graph expressing a space division model of the area.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an embodiment of the present invention, CT (Comp
The encoding of three-dimensional data obtained from uter tomography will be described with reference to the drawings.

FIG. 1 is a block diagram showing a schematic configuration of this embodiment. Reference numeral 1 denotes a data input unit, 2 denotes a region division unit of input data, and 3 denotes a space division model of a data region division result. A data structuring unit that generates and assigns a region attribute to each node of a graph expressing the same, 4 is a primitive size control unit for performing coding inside the region, and 5 is a graph creator that expresses the space division model. An area position search unit for obtaining position information of a necessary portion of data based on the search, 6 is an information processing unit for processing internal information of the area, and 7 is adapted to the attribute of the area in each area based on the search result of the graph. Area internal information encoding unit for encoding the internal information in the
Is a region position information encoding unit that searches each node of the graph expressing the space division model obtained by the data structuring unit 3 in a certain order, and encodes its attribute. The respective components will be specifically described below.

The input unit 1 is a device for inputting data to be encoded or a device for accumulating previously obtained uncoded data.

The area dividing section 2 divides the area of the input CT data and takes out the contents. An example of region division will be described with reference to FIGS. FIG. 2A is a slice image of CT data, and FIG. 2B is a histogram of CT values obtained from the entire CT data. Three-dimensional data is constructed by stacking a large number of slices as shown in FIG. The size of the data used in this example is 128 × 128
× 128.

The contents included in the CT data, for example, bones and skin, have CT values corresponding to the respective contents. Therefore, CT data area division can be performed through section division of the CT value histogram.

FIG. 3 shows the result.
(a) shows the result of histogram division. In this case, the histogram is expressed as a superposition of a plurality of functions.
Is divided into two. The division result of this histogram is
N.Ichimura: "Volume Data Coding Based on Region Seg
mentation Using Finite Mixture Model, "Proc.IEEE I
CIP-96, Vol. III, pp. 363-366, 1996.

FIG. 3B shows a slice image in which the bone and skin regions are expressed in white and gray, respectively, based on the segmentation results. By comparing with the slice image of FIG. 2A, it can be seen that the area has been divided.

The data structuring unit 3 constructs a space division model representing a region obtained as a result of the region division. The space division model is obtained as a result of recursively dividing a multidimensional space using primitives having a specific shape. Many space division models can be considered depending on the primitive used and the division method. Here, a space division model called an octree is taken as an example.

For the octree, for example, HHCh
en and TSHuang: "A Survey of Construction and Man
ipulation of Octrees, "Computer Vision, Graphics, an
The details are described in d Image Processing, Vol. 43, pp. 409-431, 1988.

The octree recursively divides a three-dimensional region using a primitive combining eight cubes. FIG. 4A shows an example. The octree can be represented by a graph as shown in FIG.

The top node of the graph represents the whole object,
Each time the graph hierarchy is increased by one, it represents a division by finer primitives. In the graph of FIG. 4B, there are three hierarchies, which correspond to the division of FIG. 4A. The top node corresponds to the entire object, and each node in the next hierarchy corresponds to each cube resulting from dividing the object into eight. The next layer corresponds to each cube resulting from further dividing each cube of the previous layer into eight cubes.

Each node of the graph is accompanied by an attribute.
In the graph of FIG. 4B, there are three attributes of Black, White, and Gray. Black and White represent the type of region. Gr
ay indicates whether the area is black or white unless it goes deeper in the hierarchy, and indicates that a plurality of attributes are mixed at this stage.

The attribute of a graph representing a conventional space division model expresses only a binary division result as described above. However, other attributes can be added in the present invention. For example, information obtained from an area division result such as an area number indicating a multi-value division result, an adjacency with another area, a distance between the area, and the like can be considered.

For example, in the case of medical data, information of the object itself such as the color, chemical property, and missing state of the tissue included in each area can be added. The information on the object itself is useful information when searching for data.

FIG. 5 is a space division model for the skull extracted as a result of dividing the CT data of FIG. 3 into regions.
Since the size of one side of the data is 128, the hierarchy of the graph representing the space division model is such that the length of one side is 128, 64, 32, 16,
8, which corresponds to a cube of size 8, 4, 2, 1. Fig. 5 (a)
Indicates that one side of the cube corresponds to 16 levels, and FIG. 5B corresponds to one level of one side of the cube. Each part of the area:
It can be easily accessed using a graph corresponding to this model at any resolution among the above-mentioned 8, 128, 64, 32, 16, 8, 4, 2, and 1. By utilizing this fact, the control of the primitive size and the search of the area position in the coding of the area internal information later are performed.

Next, the coding of the area position information will be described. The region position information encoding unit 8 in FIG. 1 encodes a graph representing a space division model. Specifically, all nodes of the graph are searched in a certain order, and a binary code is assigned to the attribute of each node.

FIG. 6 shows an example. This is an example in which the graph of FIG. 4B is coded by depth-first search.
The search order is indicated by arrows numbered 1 to 17 in FIG. Fig. 6 (b) shows the Gray node as "(" and the Black node as "
1 "and White nodes are represented by" 0 "and are arranged in search order. Fig. 6 (c) shows each symbol in Fig. 6 (b) with"("→11,"
This is an encoding result when binary codes of 1 "→ 10 and" 0 "→ 0 are assigned.After all, in this example, the space division model of FIG. It is encoded as a binary code.

As described above, the position information of the area is encoded. However, since the internal information of the area has not been encoded yet, this part will be described next. In this embodiment, the internal information of the area is a CT value in each area.

First, the primitive size control unit 4 shown in FIG.
In the following, the encoding of the internal information of the area is performed based on the primitive used in the space division model.
The reason for this is that the position of the primitive can be easily searched for in a graph representing the space division model, so that it is easy to encode only a specific part.

In the case of an octree, since the primitive is a cube, information inside each cube is encoded. In this case, it is necessary to determine the size of the cube as a unit, that is, the block size. For example, in the data of this embodiment, any one of 128, 64, 32, 16, 8, 4, 2, and 1 can be selected. Increasing the block size improves the coding efficiency, that is, increases the compression ratio, but increases the processing time. Conversely, when the block size is reduced, the processing time is shortened, but the coding efficiency is deteriorated. Therefore, the block size is given by giving priority to either the processing time or the coding efficiency.

According to the present invention, since data is hierarchically structured by the space division model, the response to the change of the primitive size changes the depth of the graph representing the space division model to be searched. There is a feature that you just do. Taking advantage of this feature, the primitive size control unit 4 can receive the primitive size determined from the calculation environment for encoding, the transmission time, etc., and change the search level of the space division model.

Next, the area position information searching section 5, information processing section 6, and area internal information encoding section 7 of FIG. 1 will be described. When encoding is performed, there is a case where only an area having a specific attribute needs to be encoded from the viewpoint of shortening calculation time and deleting unnecessary information.

According to the present invention, the data is structured by the space division model, and various attributes of each region can be attached to each node of the graph expressing the space division model. Can be easily handled.

First, the detailed processing steps of the area position searching section 5 and the area internal information encoding section 7 will be described with reference to FIG. In the description of this figure, for simplicity, it is assumed that the data of FIG. 4A is targeted, and that only the internal information of the cube corresponding to the Black node of FIG. 4B is to be encoded. Also, it is assumed that three-dimensional discrete cosine transform and three-dimensional discrete cosine transform are used for encoding the internal information.

In step 701 of FIG. 7, the search level k of the graph determined by the primitive size control unit 4 is set. In step 702, a graph search is performed as shown in FIG. In step 703, it is determined whether or not all the nodes whose attribute is Black have been searched. In step 704, it is determined whether or not the current search level is the level set in step 701. In this part, the processing is roughly divided into two.

First, the case where the current search level is not equal to k will be described. In this case, the current node is Gr
With ay, there are cases where there is a Black node at a deeper level, and that level is already White or Black. In step 705, it is determined whether the current node is Gray. If it is Gray, the process returns to step 702 to search for a deeper level in the graph. If not, it is determined in step 706 whether the node is the target Black node. Otherwise, the process returns to step 702 to continue the search.

In the case of a Black node, it means that all of the smaller cubes that make up the cube corresponding to the current node have the same attributes. In this case, first, in step 707, the area corresponding to the node is divided into blocks having the set block size. Because the current level is not k,
This is because the size of the cube corresponding to the node is larger than the set block size.

Next, in step 708, step 707
Apply a three-dimensional discrete cosine transform to all blocks divided by. Then, the resulting conversion coefficient is, for example,
Like JPEG, encoding is performed using zigzag scan, run-length encoding, and entropy encoding.

The case where the search level is equal to k in step 704 of FIG. 7 will be described. In this case, there is no need to consider the block size, and the processing is performed according to the attribute of the current node. First, in step 709, it is determined whether the node is a White node. If it is a White node, the process returns to step 702, and the search is continued. If not, it is determined in step 710 whether the node is a Black node. In the case of a Black node, it means that all the smaller cubes constituting the cube corresponding to the current node, that is, all cubes having a size equal to or smaller than the set block size have the same attribute. Therefore,
In step 711, perform uniform processing on the whole
Apply a dimensional discrete cosine transform.

On the other hand, if it is not a Black node, that is, Gr
In the case of the ay node, cubes having a size equal to or smaller than the set block size and constituting a cube corresponding to the current node do not have the same attribute. In this case as well, a three-dimensional discrete cosine transform that performs uniform processing on the whole may be applied. However, if a part other than the target area is to be coded, the coding efficiency is reduced. .

In this embodiment, as shown in step 712 of FIG. 7, a three-dimensional arbitrary shape discrete cosine transform is applied to a portion corresponding to a Gray node. FIG. 8 shows a processing procedure of the three-dimensional arbitrary shape discrete cosine transform.

In FIG. 8, the portion corresponding to the target Black node is shown in black. The block size is 8
There is. In the three-dimensional arbitrary shape discrete cosine transform, the transform is performed using a base of a one-dimensional discrete cosine transform having a length corresponding to the length of the region, such as DCT-4 in the figure. Is different from the usual three-dimensional discrete cosine transform using a base having a length equal to the block size. Using these bases, processing is sequentially performed on each axis of the block.

The transform coefficients obtained as a result of the three-dimensional arbitrary shape discrete cosine transform are also converted into a zigzag scan like JPEG, for example.
Encode using run-length encoding and entropy encoding.

As described above, when coding the internal information of a region, it is determined whether or not the region to be coded has a uniform attribute from the attributes of the nodes of the graph representing the space division model of the region. Then, based on this, it is possible to perform coding adapted to the region. This is one of the features of the present invention that is important for improving the coding efficiency.

Prior to all the steps in FIG. 7 where encoding is performed, the information is processed in the encoding target area as needed in steps 713, 714 and 715. This is effective, for example, when it is desired to emphasize and display only that part in the visualization stage.

Returning to FIG. 1, the description will be continued. The output unit 9 in FIG. 1 collects the encoding results in the area internal information encoding unit 7 and the area position information encoding unit 8 in FIG. 1 to form the final encoding result of the input data.

Next, a decoding process for reconstructing data from an encoding result will be described. FIG. 9 is a block diagram showing a schematic configuration of an embodiment of a decoding process. In FIG. 9, an encoded data input unit 91 is an input device for encoded data, or a storage device for already encoded data.

The region position information decoding unit 92 reconstructs a graph representing a space division model of the region from a binary code sequence as shown in FIG. The binary code representing the graph is
Since it is the result of searching and encoding a graph in a certain order, decoding of position information can be easily performed by reading binary codes in order and constructing nodes having attributes corresponding to the codes in the search order. I can do it.

Area position search section 5 and area internal information decoding section
In 93, an area position corresponding to the node having the attribute to be encoded is found, and its internal information is decoded.
Reconstruct the data. This process can be performed by the process shown in FIG. 7 in which encoding is replaced with decoding.

The information processing section 6 processes the reconstructed data as necessary.

The data visualizing section 94 visualizes the reconstructed data. FIG. 10 shows a result obtained by encoding and decoding the CT data shown in FIG. 2 and visualizing the result. FIG. 10 (a) shows the result of decoding and visualizing only the bone part. FIG. 10B shows the result of decoding only a part of the bone and performing information processing so that the bone is displayed translucently.

This information processing may be performed on either the sender side or the receiver side of the information, or may be performed on both sides. This result indicates that only a part of the data can be searched and encoded by the encoding method based on the data structuring.

Further, the result of FIG. 10 shows that each area is translucently displayed on the sender side and the receiver side of the information by using area search based on a graph expressing a space division model of the area. Indicates that information processing can be performed.

In the above embodiment, the present invention has been described by taking the coding of CT data as an example, but the present invention is not limited to use in coding. As another use method, an example in which data satisfying a certain attribute is searched from a database storing a large amount of encoded multidimensional data will be described below.

FIG. 11 is a block diagram showing a schematic configuration of an embodiment of a search process. In FIG. 11, an encoded data / search key input unit 111 is an apparatus for inputting encoded data and a search key from a database.

The processing in the area position information decoding section 92 and the area position search section 5 is the same as that in FIG. 9 showing the decoding process. The region attribute matching unit 112 compares the attribute of the node of the search result with the input search key. A plurality of input search keys may be used. At 113 in FIG. 11, the similarity between the two, such as the number of attributes matching the search key, is checked, and based on the similarity, it is checked whether the area corresponding to the node of the search result is the search target. If it cannot be regarded as a search target, the search is continued. If the search result can be regarded as a search result, the area is decoded by the matching area decoding unit 114, and the matching area holding unit 115
Are stored as visualization targets. Then, it is checked whether or not the entire area has been searched at 116 in FIG. If there is an unsearched area, the search is continued. If the entire area has been searched, the data in the matching area holding unit 115 is displayed in the data visualizing unit 94 as a search result.

As described above, the search from the encoded data can correspond to, for example, a search key of “clinical example having a defect in the upper right part of the skull”. This can be realized by structuring data using a space division model in encoding region position information and attaching various attributes of the region to each node of a graph representing the model.

Further, by structuring the data, it is possible to perform the attribute comparison only with the graph representing the position information of the area.
Since the decoding of the area internal information that requires the floating-point operation needs to be performed only on the portion that matches the search key as a result of the collation, efficient search can be performed. In addition, since the data is coded, the storage capacity of the database can be reduced.

[0069]

In summary, in the present invention, since the position information of the content is stored in the coded information in the form of a graph, the search for the position is replaced by the search for the graph, and the position information of the multidimensional data is obtained. The search can be performed easily and at high speed, and as a result, coding, processing, and visualization of necessary parts can be easily performed.

Further, based on the attributes of the nodes of the graph representing the space division model of the region, coding adapted to the region attributes is performed, and data retrieval from a database storing a large amount of coded multidimensional data is performed. Can be performed.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a schematic configuration of an embodiment of data encoding according to the present invention;

FIG. 2 shows multi-dimensional data (a) to be encoded;
An example of a histogram (b) of a feature used for the area division

FIG. 3 is a histogram division for area division (a).
And an example of the data area division result (b)

FIG. 4 shows an example of a space division model (a) and a graph (b) expressing the model.

FIG. 5 shows an example of a space division model of multidimensional data, in which (a) is a model in which one side of a cube corresponds to 16 layers, and (b) is a model in which one side of the cube corresponds to 1 layer.

FIGS. 6A and 6B are explanatory diagrams showing an example of encoding of area position information, in which FIG. 6A is a search order, FIG. 6B is an arrangement of attributes of each node in the search order, and FIG. In search order

FIG. 7 is a flowchart illustrating an example of processing of an area search unit, an information processing unit, and an area internal information encoding unit;

FIG. 8 is an explanatory diagram showing an example of encoding of area inside information by three-dimensional arbitrary shape discrete cosine transform;

FIG. 9 is a block diagram showing a schematic configuration of an embodiment of data decoding according to the present invention;

10A and 10B show an example of a visualization result of data obtained by decoding encoded multidimensional data, in which FIG. 10A shows a result obtained by decoding and visualizing only a bone part, and FIG. Only those that have been decrypted, processed, and visualized

FIG. 11 is a block diagram showing a schematic configuration of an embodiment when the present invention is used for data retrieval from a database;

[Explanation of symbols]

 1 is an input unit 2 is a region division unit 3 is a data structuring unit 4 is a primitive size control unit 5 is a region position search unit 6 is an information processing unit 7 is a region internal information encoding unit 8 is a region position information encoding unit 9 is an output unit 91 is an encoded data input unit 92 is an area position information decoding unit 93 is an area internal information decoding unit 94 is a data visualization unit 111 is encoded data, and a search key input unit 112 is an area attribute matching unit 114 is compatible. The region decoding unit 115 is a matching region holding unit.

Claims (4)

[Claims] 1. A method of extracting the contents of a multi-dimensional data to be encoded through region division and individually encoding the position information and the internal information of each region by encoding the position information of each region. , A space division model of a region represented by a graph is generated, a region attribute is assigned to each node of the graph, and the attribute is encoded. 2. A region attribute of each node of a graph expressing a space division model of a region includes at least information obtained from a result of the region division and information of an object included in each region.
The encoding method according to claim 1, wherein the attribute is attached as the attribute. 3. When encoding internal information of a region, a position of a coding target region is found by a search operation in a graph representing a space division model of the region, and the region is adapted in each region using the search result. The encoding method according to claim 1, wherein encoding is performed. 4. The information processing operation according to claim 1, wherein a region search based on a graph expressing a space division model of the region is used, and an information processing operation is performed on each region on a sender side and a receiver side of the information. Encoding method.