Title of Invention

A SYSTEM FOR DESCRIBING TEXTURE INFORMATION OF AN IMAGE

Abstract A texture description method in a frequency domain for extracting texture features by transforming and Gabor-filtering an input image into an image of the frequency domain, and a texture-based retrieval method thereof are provided. The texture description method in the frequency domain includes: a first step of transforming an image of a time domain into an image of the frequency domain; a second step of filtering the transformed image using a Gabor filter having NxM filter regions; a third step of extracting feature values of the image that has been Gabor-filtered in respective channels of the frequency domain division layout corresponding to the NxM filter regions of the Gabor filter; and a fourth step of describing a texture descriptor of the image on the basis of the feature values of the image. [Representative Drawing] FIG. 1 [Index Word] Texture Descriptor, Gabo Filter, Image Retrieval, Feature Vector, HVS (Human Visual System), Radon Transformation
Full Text Technical Field
The present invention relates to a system for describing texture information of an image m a frequency domain using a Gabor filter The present invention also relates to a texture description method of an image, and more particularly to a textuie description method for transforming an image of a time domain into an image of a frequency domain and extracting texture features by Gabor filtering Also, the present invention relates to a texture-based method of retrieving images indexed by the texture description method
Background Art
Texture information and its application as an indication of important \isual featuies of an image, have been studied for a long time The texture information of an image is used as a low level descriptor for content-based indexing and abstracting an image or video data Also, the texture information of the image is important in retrieving a specific photo of a digital photo album, or content-based retrieving in a tile or a textile database
Presently, feature values are calculated in a time domain or a frequency domain in order to extract texture features of an image In particular, a method of texture feature extraction in a frequency domain is known to be suitable for describing the texture features of images of a wide variety of forms
A thesis on this method, entitled "Texture Features of Browsing and Retrieval of Image Data", by B S Manjunath and W Y Ma, published on IEEE Transaction on Pattern Analysis and Machine Intelligence, Volume 18, No. 8, on August 1996, describes a method for calculating feature vectors by extracting from the image obtained after Gabor filtering in the frequency domain, the mean and the variance of each channel as feature values of the texture of an image
However, the image texture description method using the conventional Gabor filtering has problems First, it takes a long time for calculation by performing the Gabor filtering of an image in a signal domain Second, it is difficult to obtain enough information because the density of frequency samples of an image is low in the case where the texture information is extracted using the Gabor filter having a narrow pass band in a low frequency domain due to the use of an orthogonal frequency domain Third, the size
of data needed to describe features is great because both the mean and variance of an image brightness value are used as the texture features of the image
Disclosure of the Invention
To solve the above problems, it is an objective of the present invention to provide a system for describing texture information of an image in a frequency using a Gabor filter
It is another objective of the present invention to provide a texture description method and a texture-based image retrieval method in which the Gabor filtering is done in thfflfrequency domain rather than the signal domain, so as to reduce a filtering calculation time
It is yet another objective of the present invention to provide a texture description method and a texture-based image retrieval method using a Gabor filter in a frequency ddmain for extracting enough texture information of an image in a low frequency domain of the image by raising the sampling density of the frequency, and extracting enough texture information in spite of the Gabor filtering having a wide pass band by lowering the sampling density of the frequency in a high frequency domain, by performing the Gabor filtering using a polar coordinate system
It is still another objective of the present invention to provide a texture description method and a texture-based image retrieval method in a frequency domain for raising a retrieval rate using the average brightness values of an image, the variance of the brightness value of an image, energy, and a variance value of an energy for a feature descriptor of an image texture
It is one more objective of the present invention to provide, in consideration of human visual features, a texture description method and a texture-based image retrieval method using a Gabor filter in a frequency domain, by designing a frequency pass band of a Jow frequency domain to be narrower and the frequency pass band of a high frequency domain to be wider as the frequency domain becomes higher Here, it is taken into consideration that the human visual system sensitive to changes of the low frequency components of an image and insensitive to changes of the high frequency components, when designing a Gabor filter
It is one another objective of the present invention to provide a texture description
method and a texture-based image retrieval method using a Gabor filter in a frequency domain for easily retrieving an image changed by a minor amount by Gabor filtering so as not to change a value of an image texture descriptor if an image is rotated, magnified, or reduced by a small amount.
To accomplish the above objects of the present invention, there is provided a
system for describing texture information of an image in a frequency using a Gabor filter,
said system comprising
. a first means for converting/transforming an image of a time domain into an image
ofla frequency domain,
a Gabor filter having N x M filtering regions for filtering the transformed image of
the frequency domain, where N and M are predetermined positive integers,
a third means for extracting feature values of the Gabor filter image in respective
channels of a frequency domain division layout corresponding to the N x M filter regions
oflfhe Gabor filter, and
a fourth means for describing the image texture descriptor using texture feature , values of the image.
In an embodiment of the present invention, the first mean performs two-1 dimensional Fourier-transformation on the image of time domain and convert it into an image of an orthogonal coordinate system frequency domain
In another embodiment of the present invention, the first mean performs Random-transformation, and one-dimensional Fourier-transformation on the image of time domain ami converts it into an image of a polar coordinate system frequency domain
In still another embodiment of the present invention, the frequency domain division layout is obtained using a human visual system (HVS)
In yet another embodiment of the present invention, the frequency domain division layout is obtained by dividing the frequency domain by an octave interval in the radial direction away from the origin, and dividing the frequency doamin by a ' 180/division resolving power' interval in the angular direction
In still yet another embodiment of the present invention, the channels of the frequency domain division layout are granted with an importance/order of priority
In yet still another embodiment of the present invention, the third means extracts the feature values from one of the energy value and energy variance value of the Gabor-filtered image in the respective channels of the frequency domain division layout
In one more embodiment of the present invention, the texture descriptor includes energy of a DC channel, the variance of all of the pixel values of an image, the energy mean values obtained from the respective channels and/or the energy variance values obtained from the respective channels.
The present invention also provides a texture-based image retrieval system for texture-based retrieval of a data similar to a query image using a Gabor filter in a frequency domain, said system comprising
a first means for extracting a data texture descriptor by filtering data images and a mt&ns for storing the extracted data texture descriptor in a database,
a second means for extracting a query texture descriptor of a query image and a means for storing the extracted query texture descriptor;
a third means for matching the data texture descriptor and the query texture descriptor and measuring the distance between the two texture descriptors, and
a fourth means for determining a similarity between the two images according to the distance between the two texture descriptors
In an embodiment of the present invention, the first and the second means further comprise
a fifth means for transforming an image of time domain into an image of frequency domain.
a sixth means for filtering the image of the frequency domain using a Gabor filter having N X M filter regions, where N and M are respective predetermined positive integers,
a seventh means for extracting texture feature values of the Gabor-filter image in respective channels of a frequency domain division layout corresponding to the N X M filter regions of the Gabor filter, and
an eighth means for describing the texture descriptor of the image using the texture feature values of the image
In another embodiment of the present invention, the fifth mean performs tvvo-
dimensional Fourier-transformation on the image of time domain and converts it into an image of an orthogonal coordinate system frequency domain
In still another embodiment of the present invention, the fifth mean performs Rafldom-transformation, and one-dimensional Fourier-transformation on the image of time domain and converts it into an image of a polar coordinate system frequency domain
In yet another embodiment of the present invention, the frequency domain division layout is obtained using a human visual system (HVS)
In still yet another embodiment of the present invention, the frequency domain division layout is obtained by dividing the frequency domain by an octave interval in the radial direction away from the origin, and dividing the frequency domain by a '180/division resolving power' interval in the angular direction
In yet still another embodiment of the present invention, the channels of the frequency domain division layout are granted with an importance/order of priority
In one more embodiment of the present invention, the seventh means extracts the feaflure values from at least one of the energy values and energy variance values of the Gabor-filtered image in the respective channels of the frequency domain division layout
In one another embodiment of the present invention, the texture descriptor includes energy of a DC channel, the variance of all of the pixel values of an image, the energy mfc&n values obtained from the respective channels and/or the energy variance values obtained from the respective channels
In an embodiment of the present invention, the distance between the two texture descriptors is measured by respectively comparing the feature values extracted from the respective channels
In another embodiment of the present invention, the distance between the query image and the data image is measured by rotating the query image in a predetermined degree in the frequency domain, and the minimum distance is determined as the distance between two images
In still another embodiment of the present invention, the distance between the
query image and the data image is measured by size-transforming the query image in the frequency domain, and the minimum distance is determined as the distance between two images
The present invention also provides a texture description method using a Gabor filter in a frequency domain including
a first step of transforming an image of a time domain into an image of a frequency domain,
a second step of filtering the transformed frequency domain using a Gabor filter having N x M filter regions, where N and M are predetermined positive integers,
a third step of extracting texture feature values of the Gabor filtered image in respective channels of a frequency domain division layout corresponding to the N x M filter regions o-°the Gabor filter, and
a fourth step of determining a texture descriptor of the image on the basis of the teMure feature values of the image
Preferably, in the first step, the image of the time domain is Fourier-transformed into an image of the orthogonal coordinate system frequency domain, or the image of the time domain is Radon-transformed, and then one dimensional Fourier-transformed into an irflflge of a polar coordinate system frequency domain
Also, there is provided recording media which can be read by a computer in which is recorded a program for executing a texture description method using a Gabor filter in the frequency domain
Also, there is provided a texture-based image retrieval method using a Gabor filter in a frequency domain including
a first step of extracting a data texture descriptor by filtering data images using a Gabor filter and storing the extracted data texture descriptor in a data base,
a second step of extracting and storing a query texture descriptor of a query image, i e , a sought after image, when the query image is inputted using the Gabor filter,
a third step of matching the data texture descriptor and the query texture descriptor and measuring a distance between two texture descriptors, and
a fourth step of determining a similarity between two images according to the diMance between two texture descriptors
Preferably, in the first and second steps, the step of extracting the data texture descriptor and the query texture descriptor includes
a first sub-step of transforming an image of a visual domain into an image of a frequency domain,
a second sub-step of filtering the image of the frequency domain using a Gabor filter having N x M filter regions, where N and M are predetermined positive integers,
a third sub-step of extracting texture feature values of the Gabor-filteied image in respective channels of the frequency domain division layout corresponding to the N x M filter regions of the Gabor filter, and
a fourth step of describing the image texture descriptor using the texture feature values of the image
Also, there is provided recording media which can be read by a computer in which is recorded a program for executing a texture-based image retne\ al method using a Gabor filter in the frequency domain
Statement of the Invention
Accordingly the present invention relates to a system (200) for texture-based retrieval of a data similar to a query image using a Gabor filter in a frequency domain, said system comprising a system (100) for describing texture information of an image in a frequency domain using Gabor filter such as herein described, a first feature value extractor (109) that extracts a data texture descriptor by filtering data images and stores the extracted data texture descriptor in a database, a second feature value extractor (111) that for extracts a query texture descriptor of a query image and stores the extracted query texture descriptor, a matching unit (113) for matching the data texture descriptor and the queiy texture descriptor and measuring the distance between the two texture descriptors, and a determining unit (115) for determining a similarity between the two images according to the distance between the two texture descriptors
Bnef Description of the Drawings
FIG 1 is a flowchart illustrating a texture description method in the frequency domain according to a preferred embodiment of the present invention,
FIG 2 illustrates a frequency domain division layout used for extracting texture descriptors of respective channels; and
FIG 3 illustrates a structure of a Gabor filter used for extracting texture descriptors of respective channels
Fig. 4 illustrate a system for describing texture information of an in image in a frequency domain usmg a Gabor filter according to the present invention.
Best mode for carrying out the Invention
Hereinafter, "a texture description method and a texture-based image retrieval method using a Gabor filter in a frequency domain" according to a preferred embodiment of the present invention will be described in detail with reference to the attached drawings.
FIG 1 is a flowchart illustrating a texture description method in the frequency domain using a Gabor filter according to the present invention.
The texture description method, a method for producing a texture descriptor by processing an input image is used for texture information-based indexing and texture information-based retrieving an image. That is, when images, which will be indexed in the database and stored, are input, data texture descriptors are produced according to the texture description method of FIG 1, and the produced texture descriptors are stored in the

database Also, when a query image is inputted, texture descriptors of the query image are produced according to the texture description method of FIG 1, and retrieval is performed by comparing the produced texture descriptors with the data images stored in the database
Referring to FIG. 1, the texture description method according to the present invention will be described in greater detail
First, a prescribed image is inputted (SI 1), and the input image is Founer-transformed (SI2) into an image of an orthogonal coordinate system or a polar coordinate syMem frequency domain Here, the input image can be a data image or a query image as described above The input image is two-dimensional Fourier-transformed into data of the orthogonal coordinate system frequency domain Alternatively, the input image is Radon-transformed, and then the transformed data is one-dimensional Fourier-transformed into data of the polar coordinate system frequency domain
The process of transforming the input image into the polar coordinate system frequency domain is described as follows First, the input image is Radon-transformed, wherein the Radon transform is a process of line-integrating a two-dimensional image or multi-dimensional multimedia data with respect to an angle to obtain one-dimensional pflfijection data That is, the shape of an object appears to change with respect to different viewing angles, and an object is seen from all angles, so that an outline of the object can be estimated The Radon transform uses this principle
A Radon transform formula for transforming the two-dimensional image is expressed as Formula 1
(Formula Removed)
Heref(x,y) is an image function of an orthogonal coordinate system time domain, p6(R) is a first projection function obtained by linear-integrating with respect to an axis having an angle of 0 with the positive x-axis, and passing through the origin of the orthogonal coordinate system, that is, a first Radon-transform function 6(x) is a function which becomes I when x is 0 A two-dimensional image has a region of -oo x.y ooin the
orthogonal coordinate system, and regions of 0 In this way, a concurrence of the first Radon-transform function Pe(R) obtained by rotating 9 from 0 degrees through 180-degree is called a signogram The signogram is then Fourier-transformed so as to give a relation such as that shown in Formula 2 with a two-dimensional Fourier-transformed image function f(x,y) in the orthogonal coordinate system time domain
(Formula Removed) a Fourier-transformed function of the Radon-transform function pd(R). In addition, (FORMULA REMOVED)
According to the Central Slice theory, the Fourier transform of the signogram is a one-dimensional function value obtained by cutting the Fourier-transformed function of a two-dimensional original image with respect to each 6 axis If this image is Radon-transformed and then Fourier-transformed, the image is transformed into an image of the pdtar coordinate system frequency domain
Next, in a step S13, filtering is performed using the frequency domain division layout as described in FIG 2 in order to extract texture descriptors of respective channels In FIG 3, the frequency domain division layout of FIG 2 is illustrated in a more practical ftflfln Gabor filters for filtering in the step S13 can be understood to be designed on the basis of the divided feature channels as illustrated in FIG 3 That is, in the preferred embodiment, the Gabor filtering is performed using Gabor filters designed on the basis of a 5x6 feature channel spaces divided into 5 regions in the radial direction and 6 regions in the angular direction Here, respective divided frequency domains shown in FIG 2 cdfl-espond to the feature channels
The response features of the Gabor filter are expressed as Formula 3 in order to explain the operation of the Gabor filter designed as described above
(Formula Removed)
of a filter corresponding to the feature channels of the 5-th radial direction and the r-th angular direction, s indicates a position in the radial direction as an integer number among {0,1,2,3,4}, and r indicates a position in the angular direction as an integer number among {0,1,2,3,4,5} Also, Ops2 and a^2, respectively, are standard deviation values of feature channels of the coordinate p in the s direction and the coordinate 0 in the r direction, and correspond to the width of the feature channels of the radial direction and angular direction
In the Gabor filter having 5x6 filters as the preferred embodiment, exemplary stiftdard deviations of the feature channels are tabulated in Tables 1 and 2 Variable values of the Gabor filter in the radial direction are tabulated in Table 1, and variable values of the Gabor filter in the angular direction are tabulated in Table 2

(Table Removed) Table 2

(Table Removed) Next, in a step S15, the texture features of images Gabor-filtered in the orthogonal coordinate system frequency domain or in the polar coordinate system frequency domain are extracted Here, the orthogonal coordinate system frequency domain or the polar coordinate system frequency domain are divided on the basis of the human visual system (HVS) as illustrated in FIG. 2, and the divided frequency domains are each called feature channels The feature channels are indicated as C,, where 1=6 x s+r+ 7, and Co indicates a DC feature channel
The polar coordinate system frequency domain division layout divides the frequency domain on the basis of the HVS The -3 db pass band frequency features of the Gabor filter are designed to be disposed in a frequency domain to be suitable for the HVS The frequency domain division method and design principle for the Gabor filter are similarly applied to the orthogonal coordinate system. That is, a feature of the HVS is that is 1 Sensitive to the low frequency components and insensitive to the high frequency components, and the frequency division layout is decided using these features Hereinafter, this will be described in greater detail
In the present invention, the energy mean value and energy variance value of the Gafoor-filtered frequency domain are used as the texture features of the images
FIG 2 illustrates a polar coordinate frequency domain division layout for extracting the mean of energy on the basis of the HVS
As shown in FIG 2, the polar coordinate frequency domain is divided in the radial direction and the angular direction The "polar coordinate frequency domain is divided in the radial direction away from the origin at an octave interval, and in the angular direction 9 is divided into 180/P (here, P is a division resolving power of 9) If the polar coordinate frequency domain is divided like this, in the polar coordinate frequency layout to extract trfiDmean of energy, the low frequency domain is densely divided, and the high frequency domain is sparsely divided The divided frequency domains are feature channels (C), and the hatched portion is the feature channel 5
Here, important features of the present invention can be seen Sampling of the low fraquency domain is densely performed and sampling of the high frequency domain is sparsely performed by the Radon transform of the present invention, and when they are
divided on the basis of the HVS, the low frequency domain is densely divided and the high frequency domain is sparsely divided The respective divided frequency domains, that is, the feature values extracted from respective channels accurately reflect the texture features
When the energy mean value and energy variance value of the respective channels are obtained, image texture descriptors for describing an image texture from the feature values in the step SI5, that is, the feature vectors, are calculated A method for obtaining the energy mean value and energy variance value will be described later
The texture descriptors are expressed as Formula 4
(Formula Removed)
Here, e(i) is the energy mean value of the i-th Gabor-filtered channel in the frequency layout of FIG 2, and d(i) is the energy variance value of the i-th Gabor-filtered channel I in the frequency layout of FIG. 2 Here, in particular, foe indicates an energy of the DC channel, and/STD indicates a variance of all of the pixel values of the image The respective feature values of Formula 4 can be first described according to the order of priority of the channels, and the feature values of the channels having a low importance are excepted according to the importance of the channels, so that the amount of data can be reduced Also, the texture descriptor feature vectors can be formed using only an energy of the respective channels as the feature values according to the importance of the feature ofidsing both the energy and energy variance
The energy mean value (e(i)) and the energy variance value {d(i)) forming the described feature vectors are obtained by Formulas 6 and 8, and in
doing so, a valued is obtained using a Gabor-filtered function Q (zj) $) after the
Fourier-transform in Formula 5 Ifpfi) is applied to Formula 6, the energy mean value (e2)'can be obtained Also, in Formula 7, a value q(i) is obtained using the Fourier-transformed first Radon transformed function and the p(i) value obtained in Formula 5 If the q(i) is applied to Formula 8, the energy variance value (d(i)) can be obtained
(Formula Removed)
In this way, the texture descriptors formed of the energy mean value and the energy variance value of the respective channels are obtained
The step SI 1 or S16 is performed repeatedly on all input images, and the respective data texture descriptors are stored in the database
The data texture descriptors stored in the database are matched with the query texture descriptors obtained from the query image and used for retrieving images similar to thffiquery image. Hereinafter, the texture-based retrieval method according to a preferred embodiment of the present invention will be described
It is supposed that the database is indexed using the data texture descriptors (F) Next, the query texture descriptor (Fq) is extracted according to the texture description method of FIG 1, which describes the query image (q), and a conformity is measured by calculating a similarity between the texture descriptor (Fd) and the query texture descriptor (Fq) of arbitrary image data (d) within the database
This similarity is inversely proportional to the distance (Dni) between two texture descriptors obtained by Formula 9
(Formula Removed)
Here, F={f(k),k=l,K}, and The result of Fourier-transforming the rotated image is the same as the result of rotating the image in the frequency domain during the Fourier-transform of the image before the rotation Therefore, when comparing two images and retrieving, if comparing the images during the rotation in the frequency domain, two similar images having different rotation angles can be found The rotation unchangeability is expressed as Fofimula 10
(Formula Removed)
Here,

After the distance between two images is obtained by rotating the query image in the frequency domain and comparing the query image with the data image, the minimum valSie of the distance is used to indicate the amount of dissimilarity between the two final images This is expressed as Formula 11
(Formula Removed)
Also, the result of Fourier-transform of the scaled image is the same as scaling, in the frequency domain, the result of Fourier-transformation of the original image When comparing two images, if comparing the images during the scaling of the images in the frequency domain, two similar images having different sizes can be found The scale invariance is expressed as Formula 12
(Formula Removed)
Here, n is possible number of scale changes for the query image According to the preferred embodiment of the present invention, a texture-based image retrieving method includes the steps of scaling the query image in predetermined scales in the frequency damain, extracting the texture descriptor for each scale, calculating the distance between
those texture descriptors with the texture descriptor of data image, and taking the minimum distance as the distance between the two images. The step of taking the minimum distance can be expressed as Formula 13
Here, N is actual number of scaled images
(Formula Removed)
Though the present invention has been described on the basis of the preferred embodiments, the preferred embodiments are not intended to limit but to illustrate the present invention It is obvious that various changes, amendments, or controls can be made by those skilled in the art without departing from the spirit and scope of the invention. Therefore, the present invention will be defined only by the appended claims, and the above examples of changes, modifications, or controls must be understood to be included in the present invention
Referring view to Fig 4, the system (100) for describing texture information of an image in a frequency domain using a Gabor filter is illustrated by way of exemplification, said system comprising A system (100) for describing texture information of an image in a frequency domain using a Gabor filter, said system comprising.converter/transformer (101) that converts/transforms an image of a time domain into an image of a frequencv domain, the converter/transformer comprising a two-dimensional Fourier-transformation unit for determining two-dimensional Fourier-transformation on the image of time domain and converting the same into an image of an orthogonal coordinate system frequency domain,a Gabor filter (103) having N x M filtering regions for filtering the transformed image of the frequency domain, where N and M are predetermined positive integers, a feature value extractor (105) that extracts feature values of the Gabor filter image in respective channels of a frequency domain division layout corresponding to the N x M filter regions of the Gabor filtet, anda descriptor (107) that describes the image texture descriptor using texture feature values of the image
Also is shown by way of illustration a system (200) for texture based retrieval of data similar to a query image using a Gabor filter in the frequency domain, said system
comprising a system (100) for describing texture information of an image in a frequency domain, a first feature value extractor (109) that extracts a data texture descriptor by filtering data images and stores the extracted data texture descriptor in a database, a second feature value extractor (111) that for extracts a query texture descriptor of a query image and stores the extracted query texture descriptor, a matching unit (113) for matching the data texture descriptor and the query texture descriptor and measuring the distance between the two texture descriptors, and a determining unit (115) for determining a similarity between the two images according to the distance between the two texture descriptors.
Industrial Applicability
According to the present invention, not only can the texture of an image be described more accurately, but also effective indexing and retrieval becomes possible using the Gabor-filtering method in the frequency domain, the polar coordinate system frequency domain division layout suitable for extracting respective feature values, the method for extracting feature values in respective frequency domains, and the techniques for granting the importance and the order of priority to respective frequency channels
The image texture descriptor extracted by the texture description method is \ er\ useful for image retrieval when searching for an image having a special feature in a huge aerial photograph and a radar image for military purpose







WE CLAIM;
1. A system (200) for texture-based retrieval of a data similar to a query image using
a Gabor filter in a frequency domain, said system comprising:
a system (100) for describing texture information of an image in a frequency
domain using Gabor filter such as herein described;
a first feature value extractor (109) that extracts a data texture descriptor by
filtering data images and stores the extracted data texture descriptor in a database;
a second feature value extractor (111) that for extracts a query texture descriptor
of a query image and stores the extracted query texture descriptor;
a matching unit (113) for matching the data texture descriptor and the query
texture descriptor and measuring the distance between the two texture descriptors,
and
a determining unit (115) for determining a similarity between the two images
according to the distance between the two texture descriptors.
2. A system as claimed in claim 1, wherein the system (100) for describing texture
information of an image in a frequency domain using a Gabor filter comprises:
converter/transformer (101) that converts/transforms an image of a time domain
into an image of a frequency domain; the converter/transformer comprising a two-
dimensional Fourier-transformation unit for determining two-dimensional
Fourier-transformation on the image of time domain and converting the same into
an image of an orthogonal coordinate system frequency domain;
a Gabor filter (103) having N x M filtering regions for filtering the transformed image of the frequency domain, where N and M are predetermined positive integers;
a feature value extractor (105) that extracts feature values of the Gabor filter image in respective channels of a frequency domain division layout corresponding to the N x M filter regions of the Gabor filter, and
a descriptor (107) that describes the image texture descriptor using texture feature values of the image.
3. A system as claimed in claim 1 and 2, wherein the distance between the two texture descriptors is measured by the determining unit by respectively comparing the feature values extracted from the respective channels.
4. A system as claimed in claim 1 and 2, wherein the converter/transformer comprises a Random-transformation unit operatively coupled to a one-dimensional Fourier-transformation unit for converting the image of time domain into an image of a polar coordinate system frequency domain.
5. A system as claimed in claim 1 and 2, wherein the feature value extractor comprises a human visual system for obtaining the frequency domain division layout.
6. A system as claimed in claim 1 and 2, wherein the feature value extractor comprises a first dividing unit for dividing the frequency domain by an octave interval in the radial direction away from the origin, and a second dividing unit operatively coupled to the first dividing unit for dividing the frequency domain by a '180/division resolving power' interval in the angular direction thereby obtaining the frequency domain division layout.
7. A system as claimed in claim 1 and 2, wherein the distance between the query image and the data image is measured by the determining unit by size-transforming the query image in the frequency domain, and the minimum distance is determined as the distance between two images.
8. A system as claimed in claim 1 and 2, wherein the channels of the frequency domain division layout are granted with an importance/order of priority.
9. A system as claimed in claim 1 and 2, wherein the feature value extractor extracts the feature values from at least one of the energy values and energy variance values of the Gabor-filtered image in the respective channels of the frequency domain division layout.
10. A system as claimed in claim 1 and 2. wherein the texture descriptor includes
energy of a DC channel, the variance of all of the pixel values of an image, the energy mean values obtained from the respective channels and/or the energy variance values obtained from the respective channels.
11. A system as claimed in claim 1 and 2, wherein the distance between the query image and the data image is measured by the determining unit by rotating the query image in a predetermined degree in the frequency domain, and the minimum distance is determined as the distance between two images.
12. A system for describing texture information of an image in a frequency domain substantially as herein described with reference to the accompanying drawings.


Documents:

abstract.jpg

in-pct-2001-00659-del-abstract.pdf

in-pct-2001-00659-del-claims.pdf

in-pct-2001-00659-del-correspondence-others.pdf

in-pct-2001-00659-del-correspondence-po.pdf

in-pct-2001-00659-del-description (complete).pdf

in-pct-2001-00659-del-drawings.pdf

in-pct-2001-00659-del-form-1.pdf

in-pct-2001-00659-del-form-13.pdf

in-pct-2001-00659-del-form-19.pdf

in-pct-2001-00659-del-form-2.pdf

in-pct-2001-00659-del-form-26.pdf

in-pct-2001-00659-del-form-3.pdf

in-pct-2001-00659-del-form-4.pdf

in-pct-2001-00659-del-form-5.pdf

in-pct-2001-00659-del-petition 137.pdf

in-pct-2001-00659-del-petition 138.pdf


Patent Number 240122
Indian Patent Application Number IN/PCT/2001/00659/DEL
PG Journal Number 30/04/2010
Publication Date 30-Apr-2010
Grant Date 28-Apr-2010
Date of Filing 24-Jul-2001
Name of Patentee ELECTRONIC AND TELECOMMUNICATIONS RESEARCH INSTITUTE
Applicant Address 161 KAJONG-DONG, YUSONG-KU, TAEJEON, 305-350 REPUBLIC OF KOREA
Inventors:
# Inventor's Name Inventor's Address
1 KIM, MUN-CHURI 110-405 DOONGJI APT., DUNSAN -DONG, SEO-GU, TAEJON, 302-120 REPUBLIC OF KOREA
2 CHOI, YANG-LIM 102-1112 WOOMAN SUNKYUNG APT.,105 WOOMAN SUNKYUNG APT.,105 WOOMAN-DONG, PALDAL-GU, SUWON-CITY, KYUNGKI-DO, 442-190 REPUBLIC OF KOREA
3 KIM, JIN-WOONG 305-1603 EXPO APT.,JEONMIN-DONG, YUSONG-GU, TAEJON, 305-761 REPUBLIC OF KOREA
4 MANJUNATH, B.S DEPARTMENT OF ELECTRICAL AND COMPUTER ENGINEERING, UNIVERSITY OF CALIFORNIA, SANTA BARBARA, CA 93106-9560, U.S.A
5 RO, YONG-MAN 105-1503 HANWOOL APT.,SINSEONG-DONG, YUSONG-GU, TAEJON, 305-345 REPUBLIC OF KOREA
6 SHIN, YANG-DOO 510-1302 MUJIGAE MAEUL CHEONGGU APT.,221 KUMI-DONG, BUNDANG-GU, SEONGNAM-CITY, KYUNGKI-DO, 463-500 REPUBLIC OF KOREA
PCT International Classification Number G06T 7/00
PCT International Application Number PCT/KR2000/01387
PCT International Filing date 2000-11-30
PCT Conventions:
# PCT Application Number Date of Convention Priority Country
1 99-54904 1999-12-03 Republic of Korea
2 00-62260 2002-10-14 Republic of Korea