JP2001216517A - Object recognition method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the recognition precision of an object when recognizing the object in another object by volume data. SOLUTION: In an object recognition method which discriminates a voxel having a specific voxel value in volume data to discriminate a subject material in a three-dimensional object 100, an average value of voxel values of plural voxels placed in three-dimensional areas 113, 115, and 117 in the neighborhood of one voxel 111 in volume data is obtained, and a corrected voxel value of this voxel is obtained in accordance with the average value, and corrected volume data is obtained by a three-dimensional data array consisting of corrected voxel values of individual voxels in volume data, and this corrected volume data is used to discriminate the subject material inside the object.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object recognition method capable of identifying a target object inside an object expressed as volume data with high accuracy. Further, the present invention relates to image formation for displaying a perspective image of a three-dimensional structure of a target object with high contrast, and shape recognition for extracting shape information of the target object with high accuracy.

[0002]

2. Description of the Related Art With the development of image processing technology using a computer, a technology for visualizing the inside of a three-dimensional object has been utilized. Above all, in the medical field, lesions can be found at an early stage by visualizing the inside of a living body.
T (Computed Tomography) device or MRI (Magne
Medical diagnosis using a tomographic image of a living body obtained by a tic resonance imaging apparatus is widely performed.

[0003] In recent years, as a technique for obtaining a perspective image of a three-dimensional structure of a target object inside an object that is difficult to understand only with a tomographic image, volume rendering for directly drawing from three-dimensional data of the object has been attracting attention.

[0004] A minute unit area that is a constituent unit of a three-dimensional area of an object is called a voxel, and unique data representing the characteristics of the voxel is called a voxel value. The whole object is represented by a three-dimensional data array of voxel values, which is called volume data. Volume data used for volume rendering is obtained by stacking two-dimensional tomographic image data sequentially obtained along a direction perpendicular to the tomographic plane of the object. In particular, in the case of a CT image, the voxel value indicates the degree of radiation absorption at a position occupied by the voxel in the object, and is referred to as a CT value.

[0005] Ray casting is known as an excellent technique for volume rendering. Ray casting is an image image in which a three-dimensional structure inside an object is seen through the projection surface by irradiating the object with virtual light rays from a projection plane and forming an image of virtual reflected light from inside the object. Is a method of forming

[0006] An image formed on the projection plane is a virtual reflected light generated according to the opacity set for the voxel value unique to the voxel with respect to the light beam virtually illuminated from the projection plane to the voxel. These are integrated on the projection surface.

In order to obtain shape information of a target object inside the object, a specific continuous object can be extracted using volume data. By extracting voxels having the voxel value of the target object of interest from the volume data, and further extracting voxels that are in contact with each other in a voxel positional relationship from among the voxels,
Shape information of the target object of interest can be obtained. Thus, the length, diameter, volume, area, and the like of the target object can be measured.

[0008]

In volume rendering, an object structure can be drawn directly from volume data. Therefore, even in a case where a tissue such as a bone or an internal organ is complicated in a human body, the volume rendering can be performed. There is an advantage that these can be almost clearly separated and drawn.

On the other hand, the volume data obtained from the measured values has an inherent disadvantage that there is inherent superimposition noise based on the measurement principle or variation or fluctuation due to the measurement method. This drawback has the undesired consequence of reducing the contrast of the rendered image and making it difficult to identify the target object, especially when rendering internal tissues of the human body by volume rendering.

For example, when it is necessary to observe the surface of a human body tissue in detail, particularly when observing a tumor generated on the surface of an organ, the contrast of a drawn image may be reduced and the tumor may not be identified.

In the medical field, it is known to administer a contrast agent as a means for increasing the contrast of an image drawn by volume rendering. However, this puts a burden on the patient. In addition, some of the internal organs are difficult to obtain the effect of the contrast agent.

[0012] Even in the extraction of shape information of an object using volume data, the variation and fluctuation inherent in the volume data lead to erroneous voxel extraction in the extraction of voxels having the voxel value of the target object of interest. The recognition accuracy of the target object is reduced.

The present invention has been made in view of the above circumstances, and an object of the present invention is to provide an object recognition method capable of improving the accuracy of identifying a target object inside an object by volume data. Still another object of the present invention is to provide an object recognition method capable of increasing the contrast of a perspective image of a target object inside an object drawn by volume rendering. A further object of the present invention is to provide an object recognition method capable of improving the accuracy of extracting shape information of a target object inside an object using volume data.

[0014]

According to the first aspect of the present invention, there is provided an object recognition method for identifying a target object inside an object by identifying a voxel having a specific voxel value in volume data. In one of the voxels of the volume data, 3
Three-dimensional data consisting of a corrected voxel value of each voxel of the volume data, obtaining an average value of each voxel value of a plurality of voxels located in the dimension neighboring area, obtaining a corrected voxel value of the one voxel by the average value; Corrected volume data is obtained from the array, and a target object inside the object is identified using the corrected volume data.

According to the object recognition method of the first aspect,
Since the average value calculation provides a filtering effect and can suppress variations and fluctuations inherent in the volume data, it is possible to increase the accuracy of identification of a voxel having a voxel value of interest, and thereby to improve the accuracy of a target object inside an object. Accuracy of identification can be improved.

The object recognition method according to a second aspect of the present invention further comprises the step of weighting each voxel value of the plurality of voxels located in the three-dimensional neighborhood by a predetermined coefficient value. A weighted average value of voxel values of voxels is obtained, and the modified voxel value is obtained by the weighted average value instead of the average value.

According to the object recognition method of the second aspect,
By weighting according to the magnitude of the effect of the voxel in the three-dimensional neighborhood, modified volume data can be obtained more effectively.

In the object recognition method according to a third aspect of the present invention, a voxel group having a voxel value of an object of interest is extracted by converting voxel values of each voxel of the volume data into opacity. In an object recognition method for forming a perspective image of a target object inside an object by volume rendering, a perspective image is formed by using modified volume data obtained by previously performing processing according to claim 1 or 2 on volume data. Is what you do.

According to the object recognition method of the third aspect,
By forming a perspective image using the modified volume data, it is possible to suppress variations and fluctuations inherent in the volume data, and thus to enhance the contrast of the image when forming a perspective image inside the object by volume rendering. Can be. Here, the fluoroscopic image formation using the corrected volume data is to adaptively select the volume data before correction and the corrected volume data to execute the image formation, and as a result, in the image formation, In addition to using the corrected volume data, the present invention also includes a mode in which the volume data before correction is partially used and the corrected volume data is partially used.

According to a fourth aspect of the present invention, in the object recognition method, the number of the plurality of voxels located in the three-dimensional neighborhood and at least one of the coefficient values are formed by volume rendering. Is changed in accordance with the perspective image.

According to the object recognition method of the fourth aspect,
Corrected volume data can be obtained according to the perspective image of the target object inside the object to be recognized. For example, while monitoring the perspective image by ray casting, the number of voxels located in the three-dimensional neighborhood or the number of voxels The coefficient value can be variably set.

According to a fifth aspect of the present invention, there is provided an object recognition method for extracting shape information of a specific object inside an object by extracting a voxel group having a specific voxel value from volume data. The object shape information is extracted by using the modified volume data obtained by performing the processing according to claim 1 or 2 on the volume data.

According to the object recognition method of the fifth aspect,
Since the variation and fluctuation inherent in the volume data can be suppressed, the accuracy of the extracted information can be improved when extracting the shape information of the target object inside the object using the volume data.

[0024]

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

FIG. 1 shows an embodiment of the present invention. FIG. 1A shows a three-dimensional object 100 and a three-dimensional object 1.
1 shows an arbitrary cross-section 110 of FIG. 00 and an arbitrary voxel 111 of the cross-section 110. Hereinafter, for the sake of convenience, voxels are used as units, and voxel values of voxels located at coordinates (x, y, z) are represented by V (x, y, z). V
The three-dimensional array of (x, y, z) is volume data representing the object 100. Since the volume data can be coordinate-transformed for any rotation of the coordinate axes, the object 100
The following description does not lose generality even if an arbitrary cross section of is perpendicular to a specific coordinate axis.

FIG. 1B shows a voxel 111 located at coordinates (x, y, z) and a voxel near 26 of the voxel 111. The neighborhood of 26 of the voxel located at the coordinates (x, y, z) means the X coordinates (x-1) to (x +
1), Y coordinates (y-1) to (y + 1), Z coordinates (z-
This is an area obtained by removing the voxel located at the coordinates (x, y, z) from the voxels in the cubic area surrounded by 1) to (z + 1). In FIG. 1B, reference numeral 113 denotes a cross section of the voxel 111 at the X coordinate x near 26 and the voxel 1
11 and 115 and 117 indicate X near the 26, respectively.
A cross section at coordinates x-1 and x + 1 is shown.

The voxel 111 inside the object 100
, And when the object of the voxel 111 is identified from the voxel value, the voxel value V (x,
Since (y, z) is a value obtained from the measurement value, the value has inherent variations and fluctuations.

FIG. 2 shows a state of inherent variation and fluctuation inherent in voxel values. FIG. 2A shows a histogram of voxel values of an object A and an object B whose voxel values are close to each other. In FIG. 2A, 21
0 is the histograms in range v ₁ to v ₂ in which voxel values of the object A takes, 220 is a histogram of bins v ₂ to v ₃ taking the voxel values of the object B. However, the actual voxel value histogram does not become as shown in FIG. 2A because the volume data obtained from the measured values has variations and fluctuations.

FIG. 2B is obtained from actual measured values.
The histogram of the voxel values of the object A and the object B is shown.
You. 230 is a histogram of the voxel value of the object A
And 220 is a histogram of the voxel values of the object B
It is. Fluctuations and fluctuations inherent in volume data
For example, the voxel value of the object A is the original range v ₁
~ V_TwoDoes not fit. Now, paying attention to the object B, the object
Range v of voxel values of B _Two~ V_ThreeVoxel values are near
It can be seen that a voxel of the touching object A is observed.

In FIG. 1, when it is determined whether or not the voxel 111 of interest is the object B, it is verified whether or not the voxel value V (x, y, z) is in the range v _{2 to} v ₃ . However, for the above reason, there is a possibility that the voxel 111 is erroneously recognized as the object B even though the voxel 111 is the object A with a certain probability.

To solve the above problem, voxel 1
The voxel value V1 (x, y, z) is filtered as follows to obtain a modified voxel value V1
(X, y, z) is obtained. In FIG. 1B, as shown in the equation, a total of 27 voxel values obtained by adding the voxel values of the voxels in the vicinity of 26 of the voxel 111 and the voxel values of the voxel 111 itself is obtained.
, An average value of voxel values in the three-dimensional vicinity area of the voxel 111 is obtained. This average value is defined as a modified voxel value V1 (x, y, z) of the voxel 111,
By executing this process for all voxels of the object 100, corrected volume data of the object 100 can be obtained.

In the modified volume data, variations and fluctuations are suppressed by the filtering effect of the above processing as compared with volume data obtained from measured values.
The histogram of the voxel values of the objects A and B is shown in FIG.
(A), and when the voxel 111 recognizes the object B, the possibility that the voxel 111 is actually the object A is significantly reduced.

In the above-described embodiment, the description has been made using 26 neighborhoods as the three-dimensional neighborhood of the voxel of interest. However, this does not limit the setting of the three-dimensional neighborhood.
Various three-dimensional neighborhoods can be selected according to the purpose. For example, for a voxel located at coordinates (x, y, z), coordinates (x-1, y, z), (x + 1, y,
z), (x, y-1, z), (x, y + 1, z),
6 located at (x, y, z-1), (x, y, z + 1)
The six neighborhoods for selecting the voxels are an example of the smallest three-dimensional neighborhood. Or, conversely, a three-dimensional neighborhood in which the area is further enlarged than the neighborhood of 26 can be selected. In addition, it is also possible to obtain a weighted average value in consideration of the weight of each point, assuming that the influence of a point close to the point of interest in the neighboring area is large, and use this as a corrected voxel value.

FIG. 3 shows an example of a conversion function for converting each voxel value of volume data into opacity in order to perform volume rendering by ray casting. When the object A, the object B, the object C, and the object D exist inside the object 100, the conversion function of FIG.
And object D only, and 310 corresponds to an appropriate opacity to the range of the voxel value of object A, and 3
Reference numeral 20 denotes an appropriate opacity corresponding to the range of the voxel value of the object D. Since the objects B and C are not to be drawn, they are treated as transparent bodies, and the opacity is set to zero. In this way, an object to be drawn by ray casting can be extracted by the conversion function, and opacity according to the purpose of drawing can be given.

FIG. 4 is a view for explaining an embodiment of ray casting, in which a virtual ray is irradiated on the object 100 from the X-axis direction, and a virtual reflected ray reflected by each voxel is collected. ing. The reflected light beam forms an image on a projection surface (not shown) orthogonal to the reflected light beam. As described above, since the coordinate conversion of the volume data can be performed by rotating the coordinate axes, the generality is not lost by setting the direction of the virtual ray to the X-axis direction. Instead of rotating the coordinate axes, for the light beam irradiated from any direction, at the calculation point of the reflected light beam set at a pitch of voxel unit on the light beam, each voxel value is calculated according to the distance from the surrounding voxel. The same result can be obtained by performing interpolation calculation by giving a coefficient to.

In FIG. 4, 410, 420 and 430 are
X coordinates x-1, x, x + on one virtual ray respectively
Voxels located at 1 and their opacity, K
_x-1, K_x, K_{x + 1}Is given. Opacity K is 0
It is a value between 0 and 1 (0 ≦ K ≦ 1). 411, 421,
431 enters voxels 410, 420, and 430, respectively.
Ray of Light I_x-1, I_x, I_{x + 1}413, 42
3, 433 are voxels 410, 420, 430, respectively.
Ray R reflected from_x-1, R_x, R_{x + 1}It is. I
Focusing on voxel 420, the incident light I_xAgainst
And the reflected light is I_xK _xPasses through voxel 420 and
Ray I incident on voxel 430_{x + 1}Is I_x(1-K_x)
Becomes Input ray I input to voxel 420_xIs me
Opacity K of cell 420_xIs shielded based on the Shading
The ratio of the reflected light takes a value of 0 to 1,
Thus, the reflected light reflected by the voxel 420 is I_x
K_xIs multiplied by the above ratio. Reflected ray I_xK
_xRepresents the above ratio as “1”. What
Contact us for further details on the shading described later.
This will be described in terms of coefficients. From the above results, voxel 430
The reflected light of
Affected by the opacity of all voxels that passed before
You. By repeating the calculation of reflected light in this way,
When there is an object inside the object, the outer translucent object
The effect of seeing through the translucent object inside
Can see through a translucent 3D object
A fluoroscopic image can be formed.

The maximum cosine value (0-1) of the angle formed between the normal to the surface of voxel 410, 420, 430 and the direction vector of reflected light rays R _x−1 , R _x , R _{x + 1} is obtained. By multiplying the luminance, the luminance (brightness) of the surface can be obtained. Therefore, if the viewing angle changes, the cosine value changes,
The brightness changes in the range from the maximum luminance (cosine value 1) to the minimum luminance (cosine value 0). The brightness of a voxel is influenced by a plurality of voxels located near the voxel. Therefore, usually, the brightness of a certain voxel is obtained based on the cosine value of the angle between the normal line of the surface and the direction vector of the reflected light beam, which is obtained from a certain voxel and its neighboring voxels by various gradient methods. The cosine value at this time is referred to as a shading coefficient, and represents a rate at which light blocked by a certain voxel is reflected as a reflected light beam.

In the formation of a fluoroscopic image, the volume data used for ray casting is usually a measured value, and therefore inherent variations and fluctuations are inherent therein as described above, which lowers the contrast of the projected image. In the case where the voxel 420 of interest in FIG. 4 is the object A, the voxel value of the voxel 420 is as shown in FIG.
A regional portion of the histogram shown in, for example, if a value greater than v ₂ is not extracted to outside from 310 range of conversion functions of Figure 3, opacity 0 is given a transparent body As a result, the image is not drawn on the perspective image by ray casting. In order to solve this problem, one aspect of the present invention is to perform the above-described filtering process on the volume data V (x, y, z) in advance when performing ray casting,
1 (x, y, z) is obtained, and the above-described ray casting is performed using the modified volume data. In the modified volume data, due to the filtering effect,
Since the variation and the fluctuation are suppressed as compared with the volume data obtained from the measured value, the histogram of the voxel value of the object A approaches that of FIG.
20 significantly reduces the likelihood of not being extracted.

In the above-described embodiment, the three-dimensional neighboring area for filtering the volume data may be varied while observing the projected image, or the weight of each voxel of the neighboring area may be varied.

FIG. 5 shows an example of a conversion function for extracting a voxel group having a voxel value of interest in volume data. 5, 510 is a function for extracting only an object B within the object 100, the voxel value v ₂ to v ₃
Judgment value for judging existence point in voxel having
1 is given, and the other voxels are given a judgment value “0” (see FIG. 6) By applying this conversion to the volume data, voxels having the voxel value of the object B can be extracted.

FIG. 6 shows an example of a procedure for extracting a continuous object from the voxel set extracted as described above.
For convenience of explanation, FIG. 6 is shown in two dimensions. In FIG. 6, reference numerals 610, 620, 630, and 640 denote the steps of a continuous object extraction procedure in the illustrated region.
At 40, a symbol 600 indicates a voxel at the starting point of the extraction procedure, a symbol 601 indicates a voxel that is a continuation point of the next extraction procedure, and a symbol 602 indicates a voxel for which the extraction procedure processing has been completed.

Now, in step 610, processing is started from the voxel 611 indicated by the symbol 600. In the extraction of a continuous object, a voxel having a judgment value “1” for judging an existing point is selected from voxels that are in contact with a voxel currently being subjected to the extraction procedure with a surface in a three-dimensional space, and the next extraction procedure is continued. Point. In step 610, the voxel 611 at the start of the procedure first determines the judgment value "
After confirming that it is 1 ", the procedure starts.
Voxels 613, 615, and 617 that touch face-to-face with voxel 611 at the procedure start point and have determination value "1"
Is a continuation point of the next extraction procedure, and is denoted by a symbol 601. Thus, the procedure for the voxel 611 has been completed, and the symbol 602 is assigned to the voxel 611. Next, the continuous object extraction procedure is sequentially performed on the voxels that have become the continuation points of the extraction procedure.

In step 620, voxels 613 and 6
It is shown that, as a result of performing the procedures for Nos. 15 and 617, new continuation points 623, 625 and 627 of the next extraction procedure are newly obtained. Voxels 613, 615, 617
Is completed, the symbol 602 is added. Thus, the final result 640 is obtained through the step 630. 6
The set of voxels labeled 602 at 40 is a continuous object having an extracted voxel value B.

However, the volume data obtained from the measured values has inherent variations and fluctuations, and voxels other than the object B are mixed in at the stage of applying the conversion function to opacity in FIG. . As a result, errors occur in the shape information of the continuous object B extracted. According to the present invention, when the continuous object is extracted, the volume data V (x, y, z) is subjected to the above-described filtering processing in advance, and the modified volume data V1 (x,
y, z), and the above-described continuous object is extracted using the modified volume data. In the modified volume data, the filtering effect suppresses variations and fluctuations as compared with the volume data obtained from the measured values, so that the accuracy of the shape information of the extracted continuous object can be significantly improved.

[0045]

As described above, according to the present invention, in the object recognition method for identifying a target object inside an object by identifying a voxel having a specific voxel value in the volume data, For one voxel, obtain an average value of each voxel value of a plurality of voxels located in a three-dimensional vicinity area of the voxel, obtain a corrected voxel value of the one voxel by the average value, and obtain each voxel value of the voxel. Corrected volume data is obtained from a three-dimensional data array composed of corrected voxel values of voxels, and the target object inside the object is identified using the corrected volume data. By suppressing fluctuations and fluctuations, it is possible to It is possible to improve the accuracy of the identification of the object with value.

According to the present invention, there is further provided an object recognition method for forming a perspective image of a three-dimensional structure of an object of interest inside an object by ray casting. Is used, it is possible to suppress variations and fluctuations inherent in the volume data. Therefore, when forming a perspective image inside the object by volume rendering, the contrast of the image can be increased.

Furthermore, according to the present invention, in object recognition for extracting shape information of a specific target object inside an object by extracting a voxel group having a voxel value of interest from the volume data, By using the modified volume data obtained by performing the processing, it is possible to suppress the variation and fluctuation inherent in the volume data, so it is extracted when extracting the shape information of the target object inside the object using the volume data The accuracy of information can be improved.

[Brief description of the drawings]

FIG. 1 is a diagram showing a state in which an object inside an object is identified according to an embodiment of the present invention.

FIG. 2 is a diagram showing a histogram of voxel values related to identification of an object inside the object.

FIG. 3 is a view showing a conversion function for converting voxel values into opacity related to ray casting.

FIG. 4 is a diagram showing a state of performing ray casting according to an embodiment of the present invention.

FIG. 5 is a diagram showing a conversion function for converting voxel values into opacity related to extraction of an object inside the object.

FIG. 6 is a view showing a state of extracting a continuous object inside the object according to the embodiment of the present invention.

[Explanation of symbols]

100 3D object 110 Cross section of 3D object 111 Voxel 113, 115, 117 Cross section near 3D 210, 220, 230, 240 Histogram of voxel values 310, 320 Conversion function 410, 420, 430 Voxel 411, 421, 431 Incident Rays 413, 423, 433 Reflected rays 510 Conversion function 610, 620, 630, 640 One area of volume data 611, 613, 615, 617, 623, 625, 6
27 Voxel

 ────────────────────────────────────────────────── ─── Continuing on the front page (72) Inventor Kazuhiko Matsumoto 2-7-2 Araicho, Nakano-ku, Tokyo Raffine Maison Nakano B1 Zioso Software Co., Ltd. F-term (reference) 4C096 AB04 AB44 AD12 AD14 DA11 DA15 DC18 DC29 DC36 DC37 DC40 5B057 AA09 BA03 BA07 DA06 DB03 DC19 5B080 GA06 5L096 AA05 BA06 BA13 FA35 GA55

Claims (5)

[Claims] 1. An object recognition method for identifying a target object inside an object by identifying a voxel having a specific voxel value in volume data, wherein one voxel of the volume data has a three-dimensional neighborhood of the voxel. An average value of each voxel value of a plurality of voxels located in the region is obtained, a corrected voxel value of the one voxel is obtained from the average value, and a three-dimensional data array including a corrected voxel value of each voxel of the volume data is obtained. An object recognition method for obtaining corrected volume data and identifying a target object inside the object using the corrected volume data. 2. A weighted average value of the voxel values of the plurality of voxels is obtained by weighting each voxel value of the plurality of voxels located in the three-dimensional neighborhood region with a predetermined coefficient value, and 2. The object recognition method according to claim 1, wherein the modified voxel value is obtained by the weighted average value instead of the value. 3. A voxel group having a voxel value of an object of interest is extracted by converting voxel values of each voxel of the volume data into opacity, and a perspective image of a target object inside the object is formed by volume rendering. An object recognition method for forming a perspective image using corrected volume data obtained by subjecting volume data to the processing according to claim 1 or 2 in advance. 4. The apparatus according to claim 3, wherein at least one of the number of voxels located in the three-dimensional neighborhood and the coefficient value is changed according to a perspective image of an object inside the object formed by volume rendering. Object recognition method. 5. An object recognition method for extracting shape information of a specific object inside an object by extracting a voxel group having a specific voxel value from the volume data, wherein the volume data is previously stored in the volume data. An object recognition method for extracting shape information of an object using modified volume data obtained by performing the described processing.