KR100712341B1 - Method and apparatus for representing and retrieving 3d image data using 3d modified zernike moments - Google Patents

Abstract

In the present invention, the characteristics of two-dimensional or three-dimensional image data are expressed using Zernike moments, and the input query image data are discrete Fourier transformed (DFT) to an angular axis and then orthogonally transformed to a distance axis. The present invention relates to a method and apparatus for extracting a feature using an altered jerk moment, and a method and apparatus for retrieving 3D image data using the same. You can extract image features and retrieve them quickly and accurately, regardless of their rotation, size, or size.

Journey moment, Cartesian transformation, invariant, 3D object recognition, DCT

Description

TECHNICAL AND APPARATUS FOR REPRESENTING AND RETRIEVING 3D IMAGE DATA USING 3D MODIFIED ZERNIKE MOMENTS}             

1 is a flowchart of an embodiment of a feature extraction method of 3D image data using a modified low-knick moment of the present invention;

2 is a block diagram of an embodiment of an apparatus for extracting features of 3D image data by modified low-knick moment of the present invention;

3 is a flowchart of another embodiment of a method for extracting features of 3D image data by modified low nick moment of the present invention;

4 is a block diagram of another embodiment of a feature extraction apparatus for three-dimensional image data by modified low nick moment of the present invention;

5 is a flowchart of an embodiment of a feature extraction method of 3D image data that is invariant with respect to rotation and size due to a modified low-knick moment of the present invention;

6 is a block diagram of an embodiment of an apparatus for extracting features of three-dimensional image data that is invariant with respect to rotation and size due to a modified low-knick moment of the present invention;

7 is a flowchart illustrating a 3D image retrieval method using a modified low-knee moment of the present invention;

8 is a block diagram of a 3D image retrieval apparatus according to the modified low-knee moment of the present invention.

* Description of the symbols for the main parts of the drawings *

110: discrete Fourier transform unit (DFT unit) 120; Orthogonal transformation

      410; Coordinate transformation unit 420; θ-axis DFT part

      430; φ-axis DFT portion 440; Distance axis orthogonal transformation

      610; Center point estimator 620; Spherical Coordinate System Converter

      630; Normalization unit 640; 3D Journey Converter

      810; Feature extraction unit 810 'of the query image; Feature Extraction Unit

      820; A database builder 830; Video feature comparison unit

840; A comparison result output unit 850; Multimedia database

The present invention relates to a method and apparatus for extracting and retrieving features of 3D image data by a modified jockey moment, and more particularly, to extract features of an image regardless of the movement, rotation, and size of an object and to query the multimedia data. A method and apparatus for feature extraction and retrieval of three-dimensional image data capable of retrieving images quickly and accurately.

At present, the continuous development of data compression technology for storage media and delivery media, as well as the performance of presentation media and delivery media and their operating systems, allows the use of large-capacity multimedia data consisting of a plurality of mono media rather than mono media consisting of small single media. This is becoming common. Accordingly, the search for data has come to the need for multimedia data retrieval consisting of still images, voices, videos, music, etc., as well as texts, as well as conventional language-based search. Therefore, the development of an effective multimedia data retrieval method and apparatus that can easily retrieve the multimedia data desired by the user from the vast data such as the Internet or the multimedia database has been actively developed.

In the case of text data, several key words or expressions in the document can be indexed and searched. In the case of multimedia data, the multimedia data itself is large because the data itself is large in size and contains various types of information such as images, sounds, and characters. It is almost impossible to retrieve the desired multimedia data using. Therefore, in order to search for multimedia data, when constructing a multimedia database, a search must be performed by extracting a feature capable of expressing each multimedia data through a preprocessing process and comparing the extracted features with each other.

For effective multimedia data retrieval, small size of descriptors expressing the characteristics of each media data, simplification and real-time of preprocessing process, validity of information attribute expressing characteristics and flexibility of retrieval are required. Which parameter is used as the descriptor of the image feature greatly determines the efficiency of the search. In the case of the image, the features such as color, shape, texture, and motion of the image are extracted. Similarity is extracted and searched.

Conventional methods for extracting features of three-dimensional image data are roughly divided into two types, one based on surface representation and the other based on volumetric representation. Surface representation is a method of expressing features such as surface area and curvature in order to express three-dimensional objects. This method is possible under the assumption that the surface of the object is divided. However, the general object has a disadvantage that it is difficult to apply because of difficult division, and also has the disadvantage that the shape or feature of the divided surface is greatly changed according to the deformation of the object.

The volume-based method, on the other hand, is a method of obtaining the size or moment of a three-dimensional volume. This method has some strong characteristics against noise and deformation, but it has a disadvantage in that the ability to distinguish objects is inferior because there are not many feature values that can be expressed.

In addition to the above two methods, there are a method of obtaining a three-dimensional eigenvector, but this method has a problem that the orientation of the eigenvector is unstable when the object is symmetric, which makes it difficult to recognize or search by object representation. In addition, as one of the descriptors showing the characteristics of the image, a low nick moment having an invariant characteristic with respect to rotation is used. However, the existing low-knick moment is affected by the movement of the object or the change in size, and in particular, the existing two-dimensional low-knick moment is difficult to expand in three dimensions, there is a problem with a large amount of calculation.

One object of the present invention is to overcome the problems of the prior art described above, which is characterized by the use of a modified three-dimensional low-knick moment of improved performance as one of the descriptors that represent the characteristics of the image, the position, size, And a feature extraction method and apparatus for three-dimensional image data capable of effectively extracting a feature of an image regardless of rotation.

Another object of the present invention is an image retrieval method capable of quickly and effectively retrieving three-dimensional image data by extracting feature values of an image invariant to the movement, size, and rotation of an object using the modified three-dimensional low-knick moment features And to provide an apparatus.

A first aspect of the present invention for achieving the above object is to perform a discrete Fourier transform (DFT) of the input query image data to the angular axis in expressing the characteristics of the two-dimensional or three-dimensional image data by using the joke moment A method for extracting features of an image using a modified low-knee moment, comprising: obtaining an absolute value of the low-knee moment by performing orthogonal transformation on a distance axis, and a device capable of implementing the method.

A second aspect of the present invention for achieving the above object is an image information descriptor (descriptor) of the modified low-knee moment obtained by receiving the data of the query image, the discrete Fourier transform (DFT) to the angular axis and orthogonal transformation to the distance axis Feature extraction of the query image to extract the image feature using;

Another feature extraction step of extracting a feature of each image data in an image database by the same process as the feature extraction step;

An image database construction step of constructing an image database using feature extracted image data in the previous step:

Comparing the similarity between the features of the input query image and the features of the images in the image database; And

And a method for retrieving an image using a modified low-knee moment, comprising: a comparison result outputting step of outputting the similarity comparison result of the previous step.

A third aspect of the present invention for achieving the above object comprises the steps of extracting the center point of the input object from the coordinate value of the three-dimensional object of the input query image;

Converting the coordinate system into a spherical coordinate system using the center point extracted in the previous step as the origin of the coordinate;                         

Normalizing the size of the object by finding a volume in a circle around an origin and resampling it into a circle having a radius that is any percentage of the total volume;

Extracting the three-dimensional object function obtained by transforming the three-dimensional object function into the modified three-dimensional low-knick moment, and extracting it as a feature of the image data. Method and apparatus capable of implementing such a method.

A fourth aspect of the present invention for achieving the above object comprises the steps of extracting the center point of the input object from the coordinate values of the three-dimensional object by receiving the data of the three-dimensional query image;

Converting the coordinate system into a spherical coordinate system using the center point extracted in the previous step as the origin of the coordinate;

Normalizing the size of the object by finding a volume in a circle around an origin and resampling it into a circle having a radius that is any percentage of the total volume; And

Extracting the feature of the query image by converting the obtained 3D object function into the modified 3D low-knick moment and using the image information descriptor of the corresponding image to extract the feature of the 3D image data;

Another feature extraction step of extracting a feature of each image data in an image database by the same process as the feature extraction step of the query image;

An image database construction step of constructing an image database using the image data extracted from the feature in the previous step;                         

Comparing the similarity between the features of the input query image and the features of the images in the image database; And

And a method of retrieving three-dimensional image data using a modified three-dimensional low-knick moment, characterized in that it comprises a comparison result output step of outputting the similarity comparison result of the previous step.

Hereinafter, the present invention will be described in more detail with reference to the accompanying drawings.

Prior to the detailed description of the present invention, terms used in the present invention are defined as follows. Other technical terms not described in the following definitions should be construed in the ordinary meaning understood by those skilled in the art.

Definition

In the present invention, "object" refers to a specific region that the user is interested in or wants to search out of the entire image, and a part other than the object is called a background. In general object recognition, an object exists. The part can be expressed as 1, and the background part that is not an object can be expressed as 0.

In the present invention, the term "characteristic of an image" means a characteristic of color, shape, texture, motion, and the like of an image, and preferably means a texture characteristic of an image.

One aspect of the present invention, as shown in Figure 1, in representing the characteristics of the two-dimensional image data or three-dimensional image data by using the jerk moment, discrete Fourier transform the input query image data to the angular axis And extracting the absolute value of the low nick moment by performing orthogonal transformation (S2) on the distance axis (S1), and extracting the feature of the image using the modified low nick moment. The feature extraction method of the image can be preferably used to extract the feature of the texture (txture) of the image.

First, in the present invention, a discrete Fourier transform (DFT) (S1) with an angular axis so as to extract a feature of an image regardless of movement and inversion, DFT converts a one-dimensional time plane into a one-dimensional frequency plane, or two-dimensional It has the characteristic of separating the spatial plane signal into the two-dimensional frequency plane. Such a DFT is known as a transform that can divide each frequency plane evenly and effectively divide each frequency. Preferably, the Discrete Fourier Transform (DFT) of the query image data inputted in the present invention to the angular axis may include a Discrete Fourier Transform to theta (θ) axis and Discrete Fourier Transform to the pi (φ) axis. Can be.

Zernike moments are known as transformations that are independent of rotation. The basis function for the low nick moment value is

Where Rnm (ρ) is

Is defined. Where R _{n, -m} () is equal to R _nm (ρ). The jerk moment by this basic function is

Is defined as

의 저니크 모멘트 값이 A_nm 이라면 회전된 신호Is defined. Any signal

Rotated signal if the value of the low-nickel moment of A _nm is

이 된다. 결과적으로 절대값을 구할 경우 하기 식과 같이 회전된 신호의 저니크 모멘트의 절대값(Z_mn)은 회전되기 전의 신호의 그값과 같아지게 되므로(즉,

), 저니크 모멘트 값의 절대값은 회전에 관계없는 특징을 갖게 된다. The jerk moment of

Becomes As a result, when the absolute value is obtained, the absolute value (Z _mn ) of the low-knick moment of the rotated signal becomes the same as that of the signal before being rotated as follows.

), The absolute value of the low nick moment value has a characteristic independent of rotation.

The existing low-knick moments are analyzed to form a separation between the rotation axis (θ) and the distance axis (ρ), and the rotation axis has a DFT shape, and the distance axis is composed of a polynomial with Lidslow. In the present invention, by using the DFT as it is and replacing the distance axis ρ with another existing orthogonal transformation, various functions and performances can be improved, and the amount of calculation can be greatly reduced. The modified basis function for the case of using the DCT transform as an orthogonal transform method having such a characteristic can be expressed as follows.

Where k represents an arbitrary constant. When ρ has a value from 0 to 1, k can be 2π. In the present invention, in addition to the DCT, orthogonal functions such as DST, wavelet transform, Harr transform, and Fourier transform may be used as the orthogonal transform of the distance axis.

In addition, the conventional method was a two-dimensional low-kine transform of two-dimensional data, but when it is extended to three-dimensional, the rectangular coordinate system uses various orthogonal transformations with respect to the distance axis, and two angle axes (θ axis and φ axis). Using the DFT form, three-dimensional objects can be represented regardless of rotation. Extending the modified low-knick moment basis to three dimensions results in a three-dimensional low-knick moment, which will be invariant to rotation. As a result, the basis function of the modified three-dimensional low-knick moment of the present invention is                     

Is defined. Here, as the basis of distance axis MRn (ρ), as in the case of two-dimensional low-nickel moment, in addition to the polynomial, the orthogonal cos and sin functions can be used. have. If you use the cos function as an orthogonal function,

delete

It can be expressed as.

When performing an image search using the feature extraction method of the image, the query image is extracted by using data of the query image and using the electric transformed low-knee moment as an image information descriptor. In addition, a feature of each image data in the image database is extracted by the same process as the feature extraction step. The image data thus extracted is stored in the image database for use in future retrieval. During the search, the similarity between the features of the input query image and the features of the images in the image database may be compared to output the images with the highest similarity, or some related images may be output in the order of high similarity. In the latter case, the user can review and retrieve or reject the multiple images. In the present invention, the method of comparing the similarity between the image feature of the query image and the feature of the images in the image database and the method of outputting the comparison result are not particularly limited.

FIG. 2 illustrates an apparatus capable of extracting feature values that are invariant in rotation by applying the modified low-knee moment to three-dimensional image data. As shown in FIG. 2, the apparatus for extracting features of the image using the modified low-knee moment of the present invention includes a discrete Fourier transform unit (DFT unit; 110) and a discrete Fourier that receive the input query image data and FT the angular axis. And an orthogonal transform unit 120 that receives the processing result of the transform unit 110 and orthogonally transforms the distance axis to extract the absolute value of the low nick moment as an image feature descriptor. Here, the orthogonal transform unit 120 may be implemented by any one of a DFT transform unit, a DCT transform unit, a DST transform unit, a wavelet transform unit, a Harr transform unit, a Fourier transform unit, and an orthogonal polynomial transform unit. The DFT unit 110 may include a theta axis DFT unit for discrete Fourier transforming the input query image data into theta (θ) axis and a pie axis DFT unit for discrete Fourier transforming into the pi (φ) axis. The feature extraction apparatus of the image may be used as an apparatus for extracting the feature of the texture of the image.

Although not shown, the image retrieval apparatus using the modified low-knick moment of the present embodiment outputs a feature extractor of a query image, a feature extractor for constructing an image database, an image database constructor, an image feature comparer, and a comparison result output. It is configured to include a wealth. In the device of this embodiment, the query image feature extracting unit receives the query image data, transforms the discrete Fourier to the angular axis, and orthogonally transforms it to the distance axis, thereby extracting the absolute value of the modified low nick moment as the descriptor of the corresponding image. . The feature extractor extracts a feature of each image data in the image database by the same operation as that of the query image feature extractor. The image database construction unit stores the extracted feature image data to construct an image database to be searched. The image feature comparison unit compares the similarity between the characteristics of the input query image, that is, the modified low-knee moment absolute value and the modified low-knee moment absolute values of the images in the image database, so that one image or searcher having the highest similarity A group of images with high similarity is sent to the output of the comparison result for the selection. The comparison result output unit outputs the similarity comparison result received from the image feature comparison unit of the previous stage.

In another aspect of the present invention, the method for extracting features of an image may further include converting and inputting the input query image data into a spherical coordinate system before converting the input query image data into discrete Fourier transforms on an angular axis. The three-dimensional object f (x, y, z) on the Cartesian coordinate is expressed as f (ρ, θ, φ) using the spherical coordinate system ρ, θ, φ. As a result, the modified three-dimensional moment of the spherical coordinate transformed object f (ρ, θ, φ) is defined as

delete

Rotation in space of a three-dimensional object exhibits the same characteristics as the parallel movement between the θ axis and the φ axis in the spherical coordinate systems ρ, θ, and φ. As a result, the modified three-dimensional moment has a DFT-shaped change in the θ axis and the φ axis, and thus has a characteristic independent of parallel movement. As a result, it becomes a feature independent of the rotation of the original signal. For the modified three-dimensional transformation, first, the input three-dimensional signal is converted into a spherical coordinate system, and then one-dimensional transformation is performed three times on each axis, and as a result, the modified three-dimensional low nick moment can be obtained. The above equation is a change expression for analog signal, and when it is converted into a function of digital form, it is expressed as follows.

delete

According to the above equation, the absolute value of the deformed three-dimensional jerk moment is invariant to rotation.

In the method for extracting features of an image according to another aspect of the present invention, as shown in FIG. 3, after converting an input signal expressed in rectangular coordinates into spherical coordinates (S31), and then DFTing the θ axis (S32), DFT is performed on the φ axis (S33). Then, by orthogonally transforming the distance axis (S34) to obtain the absolute value of the deformed three-dimensional low-knee moment effectively extract the features of the image. The image retrieval method using the feature extraction method of the image is the same as described above, and the orthogonal transformation method that can be used is the same.

As shown in FIG. 4, an apparatus capable of implementing the image feature extraction method includes a rectangular coordinate converter 410 for receiving an input query image signal expressed in Cartesian coordinates and converting it into spherical coordinates. The Fourier transform comprises a θ-axis DFT unit 420, a φ-axis DFT unit 430 to discrete Fourier transform to the φ axis, and the distance axis orthogonal transformation unit 440. The distance axis orthogonal transform unit may be implemented by any one of a DFT transform unit, a DCT transform unit, a DST transform unit, a wavelet transform unit, a Har transform unit, a Fourier transform unit, and an orthogonal polynomial transform unit.

In another aspect of the present invention, by converting the input signal into a spherical coordinate system instead of converting the coordinates of the conversion equation to the coordinate substitution can be obtained by the following equation (1).

As a result, the modified three-dimensional low-knee moment can be obtained using the rectangular coordinate system of x, y, and z without changing the input signal into spherical coordinates. Accordingly, the apparatus for extracting features of images according to the present embodiment receives a query image signal expressed in Cartesian coordinates, and coordinates are substituted by Equation 1, θ-axis DFT unit, φ-axis DFT unit, and distance. It can be configured as an axis orthogonal transformation unit.

The modified low-knee moment of the present invention described above improves the performance of the conventional simple low-knick moment to provide a means for quickly and accurately searching the image regardless of the rotation of the input image. However, the jerk moment itself has a feature that is independent of rotation but has a weakness that is affected by movement and size change. Accordingly, another aspect of the present invention provides a method for extracting features of three-dimensional image data that is independent of the position, size, and rotation of an object, to compensate for this weakness of the low nick moment, and in addition to the above-described steps, the center point extraction step and normalization. Characterized in that the step further comprises.

Specifically, referring to FIG. 5, in this aspect, a center point extraction process (S51) is a process of determining a center point of an object to which a feature is to be extracted for feature extraction of 3D image data irrespective of the position of the object. In the present invention, the center point extraction method is not particularly limited. One preferred center point extraction method in the present invention is to determine the center of gravity of the object, that is, the first moment value as the center point. After extracting the center point, the origin of the coordinate of the input object is changed according to this value, and the coordinate system is converted into a spherical coordinate system according to Equation 2 below (S52).

f (xx ₀ , yy ₀ , zz ₀ ) = f _p (ρ, θ, φ)
Where xx ₀ = ρcosφcosθ,
yy ₀ = ρcosφsinθ,
zz ₀ = ρsinφ

delete

delete

delete

Since the converted f _p (ρ, θ, φ) is moved to the central axis, a function independent of the coordinates of the input object can be obtained.

Furthermore, the method of the present invention performs normalization to correct the size of the object in order to obtain a feature irrespective of the change in the size of the object (S53). To do this, resampling is performed to find a circle with a radius that is a constant percentage (preferably 80 to 90%) of the total volume, while the volume is rounded around the origin. This normalization process can be expressed by Equation 3 below.

delete

delete

delete

In the above formula, f _PN is a normalized three-dimensional object,

                    N is a constant representing the maximum size of a normalized object,

R is a real number between 0 and 1.0.

The 3D Jurnik moment of the normalized 3D object function is obtained up to nth order, and the feature is extracted using the 3D Jurnik moment invariant with respect to the position, size, and rotation as the feature value of the object. S54). In the present invention, for the three-dimensional low-nike transform, as described above, the orthogonal function is used as one of the orthogonal transform function of cos transform, sin transform, or DFT transform, and the complex orthogonal function of DFT transform on two angular axes is used. Can be.

6 is a block diagram of a feature extraction apparatus 600 for three-dimensional image data using the three-dimensional low-knick moment of the present invention. The feature extraction apparatus 600 of the 3D image data of the present invention includes a center point extractor 610 of a 3D object; A spherical coordinate system converter 620 for converting a rectangular coordinate system into a spherical coordinate system using the center point extracted by the center point extractor 610 as the origin of the coordinates; A normalizer 630 for normalizing the size by resampling to a circle having a radius that is an arbitrary percentage (preferably 80 to 90%) of the total volume while obtaining a volume in a circle around the origin; The three-dimensional low-knit transform unit 640 converts the obtained three-dimensional object function into three-dimensional low-knick moment and extracts it as a feature of the image data.

In the feature extraction apparatus 600 of the 3D image data of the present invention, the center point extractor 610 extracts the center point of the object from which the feature is to be extracted. In the present invention, the center point extractor is not particularly limited. It may be a means for determining the center of gravity of the center of gravity.                     

The spherical coordinate system conversion unit 620 described above is a means for converting the center point extracted by the center point extraction unit 610 into the origin of the coordinates and converting the rectangular coordinate value into the spherical coordinate value according to Equation 2 below.

[Equation 2]

f (xx ₀ , yy ₀ , zz ₀ ) = f _p (ρ, θ, φ)

Where xx ₀ = ρcosφcosθ,

yy ₀ = ρcosφsinθ,

zz ₀ = ρsinφ

Meanwhile, in the present invention, the spherical coordinate conversion unit 620 may be replaced by means for converting the coordinates of the conversion equation by coordinate substitution instead of converting the input signal into the spherical coordinate system.

[Equation 1]

delete

The normalization unit 630 normalizes the size of the object according to Equation 3 to extract the feature regardless of the change in the size of the object.                     

[Equation 3]

delete

delete

delete

delete

In the above formula, f _PN is a normalized three-dimensional object,

             N is a constant representing the maximum size of a normalized object,

R is a real number between 0 and 1.0.

Referring to FIG. 7, when performing an image search using the feature extraction method of the image, the image feature is extracted by using the data of the 3D query image and using the electric 3D low-knick moment as an image information descriptor. (S710). In other words, the descriptor of the input query image is extracted by center point extraction, transformation to spherical coordinates, normalization, and three-dimensional low-nike transformation. Meanwhile, a feature of each image data in the image database is extracted in the same process as the feature extraction step S710 (S720), and an image database is constructed from the feature extracted image data (S730). Subsequently, the similarity between the features of the input query image and the features of the images in the image database is compared (S740), and the similarity comparison result is output (S750).

Referring to FIG. 8, the feature extractor 810 of the query image receives data of a 3D query image in the apparatus for retrieving an image using the modified low-knee moment of the present invention, and receives the center point of the input object from the coordinate values of the 3D object. Center point extraction unit for extracting; A spherical coordinate system converter for converting a coordinate system into a spherical coordinate system using the center point extracted by the front end as the origin of the coordinates; A normalizer for normalizing the size of the object by resampling a circle having a radius of an arbitrary (a) percentage of the total volume while obtaining a volume in a circle around the origin; And a three-dimensional low-knit transform unit for converting the three-dimensional object function obtained in this way into a three-dimensional low-knick moment and extracting the features of the three-dimensional image data using the image information descriptor of the corresponding image. In the device the a percentage is preferably 80 to 90%. The other feature extractor 810 ′ is configured in the same manner as the query image feature extractor 810 to extract features of each image data in the image database according to the same procedure. The image database builder 820 stores the image data extracted by the feature extractor 810 ′ to construct the image database 850 so that the image extractor 810 ′ can be used immediately. The image feature comparison unit 830 compares the similarity between the features of the input query image and the features of the images in the image database, and the comparison result output unit 840 compares the similarity comparison result of the image feature comparison unit 830. Outputs In the image retrieval method and apparatus of the present invention, the feature of the image, that is, the method of comparing the similarity of the absolute value of the modified three-dimensional low-knee moment and the output method are not limited.

In the feature extraction and retrieval of the 3D image data of the present invention, other feature values may be used together to more accurately and precisely extract feature retrieval and retrieval in addition to the above-mentioned low nick moment feature values.

In the present invention, another conventional orthogonal transformation function may be used as the distance axis of the low nick moment. For example, when the DFT function is used, the size information on the distance axis is survived after conversion, and thus the size may be considered. . This DFT transformation can be done in two axes, theta (θ) axis and pi (φ) axis. DCT or DST can be used as an orthogonal transform function to create a modified jerk that can concentrate energy at low frequencies. In addition, the distance-axis orthogonal transformation can use the Har, Fourier, wavelet, or orthogonal polynomial.

In the present invention, the low-knick moment extended in three dimensions is invariant to rotation, and furthermore, exhibits strong characteristics such as deformation, shape, movement, and size of the object, and also has strong characteristics against noise. It provides a means for feature extraction and retrieval of the 3D image data.

Claims (50)

In expressing the characteristics of two-dimensional or three-dimensional image data by using the low-knick moment, the process of obtaining the absolute value of the modified low-knick moment by performing discrete Fourier transform (DFT) on the angular axis and orthogonal transformation on the distance axis Method for extracting features of the image using the modified low-knee moment, characterized in that it comprises a. The method of claim 1, further comprising converting the image data into a spherical coordinate system and expressing the image data before converting the image data into discrete Fourier transforms on an angular axis. Extraction method. The modified method of claim 1, wherein the method further comprises the step of performing coordinate substitution of the rectangular coordinate system of the image data by Equation 1 before the discrete Fourier transform (DFT) of the image data. Feature Extraction Method of Image Using Low Nickel Moment. [Equation 1]

The method of claim 1, wherein the discrete Fourier transform (DFT) of the image data comprises the steps of discrete Fourier transform to theta (θ) axis and discrete Fourier transform to the pi (φ) axis A feature extraction method of an image using a modified low nick moment. 2. The method of claim 1, further comprising the step of converting the image data into a spherical coordinate system and presenting the discrete data using discrete Fourier transform on the angular axis, the method further comprising discrete Fourier transform (DFT) on the angular axis. The method includes extracting discrete Fourier transform into theta (θ) axis and Discrete Fourier transform into pi (φ) axis. The image of the query image is extracted by using the deformed low-knee moment absolute value obtained by performing discrete Fourier transform (DFT) on the angular axis and orthogonal transformation on the distance axis as the image information descriptor. Feature extraction step; A feature extraction step of extracting a feature of each image data in an image database by the same process as the query image feature extraction step; An image database building step of constructing an image database using the feature extracted image data: Comparing the similarity between the features of the input query image and the features of the images in the image database; And And a comparison result outputting step of outputting the similarity comparison result of the previous step. 7. The modified Jurnik according to claim 6, wherein the method further comprises converting the input query image data into a rectangular coordinate system before converting the input query image data into discrete Fourier transforms on an angular axis. Search method of image using moment. 7. The method of claim 6, further comprising the step of performing coordinate substitution of the rectangular coordinate system of the input query image data by Equation 1 before the discrete Fourier transform (DFT) of the input query image data to the angular axis. Image retrieval method using the modified low-knee moment, characterized in that. [Equation 1]

7. The method of claim 6, wherein the discrete Fourier transform (DFT) of the input query image data comprises an Fourier transform of the input query image data to an θ axis and a Discrete Fourier transform to a pi (φ) axis. Image retrieval method using a modified low-knee moment, characterized in that. 2. The method of claim 1, further comprising the step of converting the image data into a spherical coordinate system and presenting the discrete data using discrete Fourier transform on the angular axis, the method further comprising discrete Fourier transform (DFT) on the angular axis. The method includes extracting discrete Fourier transform into theta (θ) axis and Discrete Fourier transform into pi (φ) axis. The method according to any one of claims 1 to 5, wherein DCT is used as an orthogonal transformation of the distance axis. The method according to any one of claims 1 to 5, wherein a DST is used as an orthogonal transformation of the distance axis. The method according to any one of claims 1 to 5, wherein a wavelet transform is used as an orthogonal transformation of the distance axis. The method according to any one of claims 1 to 5, wherein a Harar transform or a Fourier transform is used as an orthogonal transformation of the distance axis. The method according to any one of claims 1 to 5, wherein an orthogonal polynomial transformation is used as an orthogonal transformation of the distance axis. The method of claim 1, wherein the method is a method of extracting a feature of a texture of an image. In the feature extraction apparatus for two-dimensional or three-dimensional image data, A DFT unit for receiving input query image data and converting discrete Fourier to an angular axis; And an orthogonal transform unit for receiving the processing result of the discrete Fourier transform unit and orthogonally transforming the distance axis to obtain an absolute value of the modified low nick moment. 18. The apparatus of claim 17, wherein the apparatus further comprises a spherical coordinate converter for converting the query image data inputted at the front end of the DFT unit into a spherical coordinate system. 18. The method according to claim 17, wherein the apparatus further comprises means for performing coordinate substitution of an orthogonal coordinate system of the query image data inputted at the front end of the DFT unit by the following equation (1). Features Extractor. [Equation 1]

18. The apparatus of claim 17, wherein the apparatus comprises a theta axis DFT unit for discrete Fourier transforming the input query image data into theta (θ) axis and a pie axis DFT unit for discrete Fourier transforming on the pi (φ) axis. An apparatus for extracting features of an image using the modified low nick moment. It consists of a DFT unit that receives the input query image data and transforms it into discrete Fourier transform on the angular axis, and an orthogonal transform unit that obtains the absolute value of the transformed low nike moment by orthogonally transforming it to the distance axis after receiving the processing result of the discrete Fourier transform. A query image feature extraction unit for extracting an absolute value of the nick moment into a descriptor of the corresponding image; Another feature extractor which extracts a feature of each image data in an image database by the same operation as the feature extractor; An image database construction unit for constructing an image database using the image data extracted from the features in the previous step: An image feature comparison unit comparing the similarity between the features of the input query image and the features of the images in the image database; And And a comparison result output unit for outputting the similarity comparison result of the previous step. 22. The apparatus of claim 21, wherein the feature extractors further include a spherical coordinate converting unit converting the query image data input at the front end of the DFT unit into a spherical coordinate system. 22. The method of claim 21, wherein the feature extractors further include a means for performing coordinate substitution of an orthogonal coordinate system of the query image data inputted at the front end of the DFT unit by the following Equation 1. Image retrieval device. [Equation 1]

22. The apparatus of claim 21, further comprising a theta axis DFT unit for discrete Fourier transforming the input query image data into theta (θ) axes and a pie axis DFT unit for discrete Fourier transformation on the pi (φ) axes. Apparatus for retrieving an image using the modified low-knee moment, characterized in that. 18. The apparatus of claim 17, wherein the orthogonal transform unit is a DFT transform unit. 21. The apparatus for extracting features of an image according to any one of claims 17 to 20, wherein the orthogonal transform unit is a DCT transform unit. 21. The apparatus of claim 17, wherein the orthogonal transform unit is a DST transform unit. The apparatus of claim 17, wherein the orthogonal transform unit is a wavelet transform unit. The apparatus of claim 17, wherein the orthogonal transform unit is a Har transform unit or a Fourier transform unit. The apparatus of claim 17, wherein the orthogonal transform unit is an orthogonal polynomial transform unit. 18. The apparatus of claim 17, wherein the apparatus is an apparatus for extracting texture features of an image. Extracting a center point of the input object from the coordinate values of the three-dimensional object of the image data; Converting the coordinate system into a spherical coordinate system using the center point extracted in the previous step as the origin of the coordinate; Normalizing the size of the object by resampling to a circle with a radius equal to any (x) percent of the total volume while finding the volume in a circle around the origin; Extracting the feature of the 3D image data by the modified 3D Jurnik moment, comprising converting the obtained 3D object function to the modified 3D Jurnik moment and extracting it as a feature of the query image Way. Receiving data of a 3D query image and extracting a center point of the input object from coordinate values of the 3D object; Converting the coordinate system into a spherical coordinate system using the center point extracted in the previous step as the origin of the coordinate; Normalizing the size of the object by resampling to a circle having a radius of any (a) percent of the total volume while finding a volume around the origin; And Extracting the feature of the query image by converting the obtained 3D object function into the modified 3D Jurnik moment and extracting it as a feature of the query image; Another feature extraction step of extracting a feature of each image data in an image database by the same process as the query image feature extraction step; An image database building step of constructing an image database using the feature extracted image data; Comparing the similarity between the features of the input query image and the features of the images in the image database; And And a comparison result outputting step of outputting the similarity comparison result of the previous step. 33. The method of claim 32, wherein the extracting the center point is extracting the center of gravity of the object as the center point. 33. The method of claim 32, wherein the converting into the rectangular coordinate system comprises converting the rectangular coordinate system into a rectangular coordinate system according to Equation 2 below. . [Equation 2] f (xx ₀ , yy ₀ , zz ₀ ) = f _p (ρ, θ, φ) Where xx ₀ = ρcosφcosθ, yy ₀ = ρcosφsinθ, zz ₀ = ρsinφ 33. The modified 3D low-order low-order storage method according to claim 32, wherein the method comprises: substituting a rectangular coordinate system of the input query image data by Equation 1 instead of converting the coordinate system into a spherical coordinate system. Feature Extraction Method of Three-Dimensional Image Data by Nick Moment. [Equation 1]

33. The method of claim 32, wherein the normalizing comprises obtaining a normalized three-dimensional object according to Equation 3 below. [Equation 3]

In the above formula, f _PN is a normalized three-dimensional object,              N is a constant representing the maximum size of a normalized object,              R is a real number between 0 and 1.0. 33. The method of claim 32, wherein the three-dimensional low-kine transform step uses an orthogonal transform function of cos transform, sin transform, or DFT transform for the genre transform, and complexes the DFT transform on two angular axes. Method for extracting the feature of the three-dimensional image data by the modified three-dimensional Journey moment, characterized by using an orthogonal function. 39. The method of claim 38, wherein the DFT transform comprises discrete Fourier transform into a theta (θ) axis and discrete Fourier transform into a pi (φ) axis. 33. The method of claim 32, wherein the method is a method of extracting a feature of a texture of an image. 33. The method of claim 32, wherein the arbitrary (a) percent is from 80 to 90 percent. A center point extracting unit extracting a center point of the input object from the coordinate values of the three-dimensional object; A spherical coordinate system converting unit converting the coordinate system into a spherical coordinate system using the center point extracted in the previous step as the origin of the coordinates; A normalizer for normalizing the size of an object by resampling a circle having a radius of an arbitrary (a) percentage of the total volume while obtaining a volume around the origin; And The three-dimensional image by the modified three-dimensional low-knee moment, characterized in that it comprises a three-dimensional low-knit transform unit for converting the three-dimensional object function obtained in this way into a modified three-dimensional low-knick moment and extracting it as a feature of the image data Data feature extraction device. A center point extracting unit which receives data of a 3D query image and extracts a center point of the input object from coordinate values of the 3D object; A spherical coordinate system converting unit converting the coordinate system into a spherical coordinate system using the center point extracted by the center point extracting unit as the origin of the coordinates; A normalizer which normalizes the size of the object by resampling a circle having a radius of an arbitrary percentage of the total volume while obtaining a volume around the origin; A feature extraction unit of the query image including a 3D low-knee transform unit for converting the obtained 3D object function into a modified 3D low-knick moment and extracting it as a feature of the corresponding query image; Another feature extractor which extracts a feature of each image data in an image database by the same operation as that of the query image feature extractor; An image database constructing unit for constructing an image database from the image data extracted by the other feature extracting unit; An image feature comparison unit comparing the similarity between the features of the input query image and the features of the images in the image database; And And a comparison result output unit configured to output the similarity comparison result of the image feature comparator. 43. The apparatus of claim 42, wherein the center point extractor is a means for extracting the center of gravity of the object as a center point. 43. The apparatus of claim 42, wherein the spherical coordinate system converter is a means for converting the spherical coordinate system into a spherical coordinate system according to Equation 2 below. [Equation 2] f (xx ₀ , yy ₀ , zz ₀ ) = f _p (ρ, θ, φ) Where xx ₀ = ρcosφcosθ, yy ₀ = ρcosφsinθ, zz ₀ = ρsinφ 43. The apparatus of claim 42, wherein the apparatus includes means for coordinate replacement of the rectangular coordinate system of the query image data input instead of the spherical coordinate system conversion unit by Equation 1 below. Feature extraction apparatus for 3D image data. [Equation 1]

43. The apparatus of claim 42, wherein the normalization unit is a means for obtaining a three-dimensional object normalized according to Equation 3 below. [Equation 3]

In the above formula, f _PN is a normalized three-dimensional object,              N is a constant representing the maximum size of a normalized object,              R is a real number between 0 and 1.0. 43. The complex quadrature of claim 42, wherein the three-dimensional low-neak transform unit uses one orthogonal transform function of cos transform, sin transform, or DFT transform for the L'Genre function for low-kine transform. And a means for utilizing a function. 3. 49. The apparatus of claim 48, wherein the apparatus is a means for performing Discrete Fourier transform to theta (θ) axis and Discrete Fourier transform to pi (?) Axis by DFT transform. 43. The apparatus of claim 42, wherein the arbitrary (a) percentage is 80 to 90 percent.

Images