• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    PURPOSE: A contents based image retrieval system and a method are provided to be capable of rapidly retrieving a desired image by editing and deforming an inquiry image input by a user or an image output by the retrieve to reuse the image in inquiry. CONSTITUTION: An image input part(100) receives images for structuring an image descriptor database. An image extraction part(101) extracts an ART(Angular Radial Transform) image descriptor of images input to the image input part(100). Images input from the image input part(100) are stored in an image database(102). Image descriptors extracted from the image extraction part(101) are stored in an image descriptor database(103). Images to be retrieved from a remote user web browser are input to an image inquiry part(104). An inquiry image extraction part(105) extracts an image descriptor from the inquiry images. An image descriptor similarity comparator(106) compares the similarity between the image descriptor of the inquiry image and the image descriptor of a database image recorded in the image descriptor database(103). An image output part(107) outputs an image similar to the inquiry image retrieved by the comparator(106).
    公开号:KR20010053788A
    申请号:KR1019990054300
    申请日:1999-12-01
    公开日:2001-07-02
    发明作者:김영섬；김용성
    申请人:송문섭；주식회사 현대큐리텔；김회율；김영섬；주식회사 코난테크놀로지；
    IPC主号:
    专利说明:

    Content-based image retrieval system and its method {SYSTEM FOR CONTENT-BASED IMAGE RETRIEVAL AND METHOD USING FOR SAME}
    [11] BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a content-based retrieval system for retrieving data using features that can be automatically extracted from multimedia data. In particular, the similarity of object shapes included in still images and videos in a network environment including the Internet is described as ART ( Content-based image retrieval system and method for retrieving images with similar shapes regardless of position, size, and rotation angle and user's query image by automatic judgment using Angular Radial Transform coefficient, and realization The present invention relates to a computer-readable recording medium having recorded thereon a program.
    [12] With the recent development of the Internet and the rapid increase in the use of digital images, conventional image retrieval using text does not guarantee reliable search results.
    [13] Accordingly, recently, a method of searching for an image through direct input of an image has been in the spotlight.
    [14] Image retrieval using images refers to similar images by extracting an image descriptor describing the characteristics of the image from the image, and comparing the similarity between the query image input by the user and the image descriptor of the image stored in the database. This is how you search. Here, image descriptors can be classified into a color descriptor describing a color of an image, a texture descriptor describing a texture, and a shape descriptor describing a shape. The efficiency of the system depends on how efficiently the image descriptors represent the characteristics of the image.
    [15] As a conventional shape descriptor, an invariant moment descriptor is mainly used, and the invariant moment descriptor has a characteristic of having an invariant value with respect to size conversion, movement, and rotation of an image.
    [16] In order to obtain the moment descriptor from the input image, the edge detection and binarization process for separating the object from the background using the spatial features of the image is first performed, and the outer boundary of the object is detected from the separated background. Get the shape vector of the object from the separated object.
    [17] Next, to measure the similarity between the input image and the image stored in the database, an Euclidean distance measuring method such as Equation 1 below is used.
    [18]
    [19] Here, q is an input image, t is an image stored in the database, H p is a moment value of the input image q, H t is a moment value of the database image t, M has a value between 0-6.
    [20] However, the content-based image retrieval system using a conventional moment descriptor as described above is a descriptor that expresses an image because there is overlap of information between the extracted moment values because the polynomial functions used as the basis functions are not orthogonal to each other. Inefficient, there is a problem that does not properly reflect the cognitive characteristics of the user does not properly retrieve the image of a similar shape felt by the user.
    [21] Therefore, the present invention devised to solve the above problems, using the absolute value of the coefficient of the ART (Angular Radial Transform) coefficient as an image descriptor for content-based image retrieval, as a result of the query image or search input by the user To provide a content-based image retrieval system and method for quickly retrieving a desired image by editing and transforming the output image and re-using the query, and a computer-readable recording medium recording a program for realizing the same. The purpose is.
    [1] 1 is a block diagram showing a schematic configuration of a content-based image retrieval system according to the present invention;
    [2] Figure 2 is an embodiment processing flow diagram for performing a content-based image retrieval method according to the present invention.
    [3] 3 is a detailed flowchart illustrating an embodiment of an image database construction process in FIG. 2;
    [4] 4 illustrates an image function set in accordance with the present invention.
    [5] 5 is an exemplary image classification of an image database according to the present invention;
    [6] * Explanation of symbols for the main parts of the drawings
    [7] 100: image input unit 101: image descriptor extraction unit
    [8] 102: Image Database 103: Image Descriptor Database
    [9] 104: image query unit 105: query image descriptor extraction unit
    [10] 106: image descriptor similarity comparison unit 107: image output unit
    [22] Content-based image retrieval system according to the present invention for achieving the above object, the image input means for inputting the image for building the database; Image descriptor extraction means for extracting an ART (Angular Radial Transform) image descriptor of the image inputted to the image input means; Image storage means for storing the image inputted from the image input means; Image descriptor storage means for storing the image descriptor extracted by the image descriptor extraction means; Image query means for inputting an image to be searched by the user; Query image descriptor extraction means for extracting an image descriptor from an input query image; Image descriptor dissimilarity measuring means for receiving an image descriptor for the query image from the query image descriptor extracting means and measuring the dissimilarity of the reference image recorded in the image descriptor storage means with the image descriptor; And image output means for receiving an image similar to the query image from the image descriptor dissimilarity measuring unit in the image database of the database server and outputting the image for the user to confirm.
    [23] Content-based image retrieval method according to the invention, the first step of building a database for content-based image retrieval; A second step of inputting a query image to be searched using the image input means; Extracting an ART image descriptor from the query image input by the image descriptor extracting means to measure dissimilarity between the image stored in the database and the image descriptor, and obtaining a similar image; And a fourth step of outputting the image retrieved from the database for the user to check.
    [24] The present invention also provides a content-based image retrieval system having a microprocessor, comprising: a first function of constructing a database for content-based image retrieval; A second function of inputting a query image to be searched using the image input means; A third function of extracting an ART image descriptor from the query image input by the image descriptor extracting unit to measure dissimilarity between the image stored in the database and the image descriptor, and obtaining a similar image; And a computer-readable recording medium having recorded thereon a program for realizing a fourth function of outputting the image retrieved from the database for the user to confirm.
    [25] The above objects, features and advantages will become more apparent from the following detailed description taken in conjunction with the accompanying drawings. Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.
    [26] 1 is a block diagram showing a schematic configuration of a content-based image retrieval system according to the present invention.
    [27] As illustrated in FIG. 1, an image input unit 100 receiving an image for constructing an image and an image descriptor database, and an image descriptor extracting unit extracting an ART image descriptor of an image input to the image input unit 100 ( 101, an image database 102 for storing an image input from the image input unit 100, an image descriptor database 103 for storing an image descriptor extracted from the image descriptor extracting unit 101, and a user An image query unit 104 for inputting an image to be searched in a remote user web browser, a query image descriptor extracting unit 105 for extracting an image descriptor from a query image input from the remote user web browser, and extracting the query image descriptor In unit 105, the image descriptor for the query image is received and the image descriptor database 1 An image descriptor similarity comparison unit 106 for determining the similarity with the image descriptor of the database image recorded in 03), and an image similar to the query image retrieved by the image descriptor similarity comparison unit 106; It is provided with an image output unit 107 for receiving so that the user can check.
    [28] In the present invention, a shape descriptor is used as an image descriptor for content-based image retrieval, and among them, an absolute value of an Angular Radial Transform (ART) coefficient is used.
    [29] ART (Angular Radial Transform) has rotational invariance, which is an essential element in image retrieval. Rotation invariant herein means that the image descriptor extracted by ART does not change even when the image is rotated.
    [30] First, the ART definition equation for obtaining the ART image descriptor is expressed by the following "Equation 2".
    [31]
    [32] Where n and m are integers that determine the order of the ART basis, Fnm is the ART coefficient, and Vnm (ρ, θ) is the ART basis function expressed in the polar coordinate system (ρ, θ), and f (ρ, θ) means image, superscript '*' means conjugate complex number)
    [33] The ART basis function Vnm (ρ, θ) of the above ART definition formula (see "Equation 2") is defined by the following "Equation 3".
    [34]
    [35] (Where A m (θ) is an Angular function constituting the ART basis function, and R n (ρ) is a Radial Function constituting the ART basis function)
    [36] Incidentally, the declination function A m (θ) of Equation 3 is again defined by Equation 4 below.
    [37]
    [38] The ART definition formula (see Equation 2) uses a (ρ, θ) polar coordinate system instead of a (x, y) rectangular coordinate system to obtain rotational invariance. The polar coordinate system is a method of expressing coordinates by an angle θ formed between a distance ρ and an x axis spaced from an origin point.
    [39] R n (θ) of Equation 3 may have various types. For example, R n (θ) of Equation 3 may be represented by ART-C as shown in Equation 5 below.
    [40] 4 is an exemplary diagram of a set of ART basis functions of the ART-C type.
    [41]
    [42] The value Fnm obtained by the above Equation 2 is a series of complex numbers. In the present invention, as shown in Equation 6 below, a vector obtained by taking an absolute value to the value of the Fnm is a shape descriptor. Is defined as "."
    [43]
    [44] (Where n = 0,1,2 ,,, k and m = 0,1,2 ,,, l)
    [45] Since the ART coefficient extracted from the image indicates how much of the components of the ART basis function are included in the original image, the combination of the product of the ART coefficient and the ART basis function can restore the original image. Theoretically, an infinite number of ART coefficients and the basis function may be combined to obtain an image exactly the same as the original image, but in practice, only 20 to 30 pieces of information may be combined to obtain an image having almost no error with the original image. 5). In other words, the image can be represented by 20 to 30 numbers, which means that the ART coefficient can be an efficient descriptor for the image.
    [46] On the other hand, ART has orthonormality as defined by Equation 7 below.
    [47]
    [48] (However, δ is a Kroneker delta function, which is 0 when n = n 'and m = m', and has a value of 1 in other cases.)
    [49] The absolute value of the ART coefficient has a rotation invariance as defined by Equation 8 below.
    [50]
    [51] This can be easily seen by extracting the ART coefficient from the image f α (ρ, θ) where the original image f (ρ, θ) is rotated by an angle α. That is, when two images f (ρ, θ) and f α (ρ, θ) have the same relationship as in Equation 8 above, the ART coefficient extracted from the image f α (ρ, θ) is Equation 9 ", wherein the relationship between the original image f (ρ, θ) and the ART coefficients extracted from the image f α (ρ, θ) rotated by the angle α is expressed by Equation 10 below.
    [52]
    [53] (Where F α nm is the ART coefficient extracted from the image f α (ρ, θ) where f (ρ, θ) is rotated by an angle α)
    [54]
    [55] (Where Fnm and F α nm are ART coefficients extracted from f (ρ, θ) and f α (ρ, θ), respectively )
    [56] However, when the absolute value of the F α nm value of the rotated image is taken, the same value as the Fnm value of the original image is obtained, as shown in Equation 11 below. have.
    [57]
    [58] Equation 12 below is an equation representing the degree of dissimilarity between the query image and the database image. That is, since descriptors extracted from similarly shaped images have similar values, and descriptors extracted from differently shaped images have completely different values, the difference between descriptors extracted from two images is expressed as shown in Equation 12 below. By comparison, you can determine how similar the two images are.
    [59]
    [60] (Where D is the dissimilarity between the query image and the database image, W i is the constant coefficient, S i q is the i-th image descriptor of the query image, and S i r is the i-th image descriptor of the database image)
    [61] Now, the content-based image retrieval method using the ART image descriptor will be described in detail.
    [62] 2 is a flowchart of an embodiment of a content-based image retrieval method according to the present invention.
    [63] First, in the process of building an image database for image retrieval, the image descriptor database 103 and the image database 102 are constructed using the image input unit 100 and the image descriptor extracting unit 101 for the database server (S200). In operation S202, the user inputs an image to be searched in the remote user web browser to the image querying unit 104. In this case, the image query unit 104 may create an image to be directly searched by the user by using a mouse or a digitizer, select an image from a prototype image provided by a web server, and It has a function of loading an image stored in a hard disk, and has a function of allowing a user to read and change an existing image.
    [64] When the user inputs a query image, the query image descriptor extracting unit 104 extracts an image descriptor of the input query image (S204) and transmits the image descriptor to the image descriptor similarity comparing unit 106.
    [65] The image descriptor similarity comparison unit 106 which receives the image descriptor of the query image from the query image descriptor extracting unit 104 compares the similarity with the image descriptor stored in the image descriptor database 103, If similar images are determined (S206), the images corresponding to the determined image descriptors are acquired from the image database 102, sorted in the order of similarity, and then transmitted to the image output unit 107 (S208). The unit 107 displays the searched image on the web browser so that the user can check the searched image (S210). Here, the retrieved image is reused as a prototype image provided by a web server, and a user may re-search by modifying a prototype image provided by a web server to a remote user web browser. In addition, the image output unit 107 provides a representative image as a prototype image to the user upon initial access of the user, and transmits the selected prototype image to the image query unit 104 when the user selects one of them. .
    [66] FIG. 3 is a detailed flowchart of a process of constructing an image descriptor database 103 and an image database 102 using the image inputter 100 and the image descriptor extracting unit 101 for the database server in FIG. 2. Input an image to be stored in the database using the image input unit 100 (S300), the image descriptor extracting unit 101 extracts an image descriptor from the input image (S302), the image descriptor database 103 Record (S304).
    [67] In addition, the image is recorded in the image database 102 using the image input unit 100 (S306).
    [68] 4 is an exemplary diagram of an image function set according to the present invention, which illustrates an ART-C type ART base function set described in Equation 5 above. In the figure, n represents the distance from the origin and m represents the angle from the x axis.
    [69] FIG. 5 is an example of content-based image retrieval according to the number of image functions according to the present invention. FIG. 5 illustrates an example of synthesizing an image similar to a query image by using a combination of ART definition expressions. It indicates the number of ART definition expressions.
    [70] As shown in the figure, as the number of combined ART definitions increases, it can be seen that images similar to query images are synthesized.
    [71] The present invention described above is capable of various substitutions, modifications, and changes without departing from the spirit of the present invention for those skilled in the art to which the present invention pertains. It is not limited to the drawing.
    [72] As described above, the present invention is an image descriptor for content-based image retrieval, which is a transform coefficient using an orthogonal basis function, thus having no information redundancy, rotation insomnia, and an efficient image descriptor that reflects human visual characteristics. Using an Angular Radial Transform (ART) descriptor, you can quickly and accurately retrieve images from the database that look similar to the query image you entered, and prototype the query image or the searched result image you entered. By reusing images, there is an excellent effect of performing finer image retrieval.
    权利要求:
    Claims (22)
    [1" claim-type="Currently amended] Image input means for inputting an image for building a database;
    Image descriptor extraction means for extracting an ART (Angular Radial Transform) image descriptor of the image inputted to the image input means;
    Image storage means for storing the image input from the image input means;
    Image descriptor storage means for storing the image descriptor extracted by the image descriptor extraction means;
    Image query means for inputting an image to be searched by the user;
    Query image descriptor extraction means for extracting an image descriptor from an input query image;
    Image descriptor dissimilarity measuring means for receiving an image descriptor for the query image from the query image descriptor extracting means and measuring the dissimilarity of the reference image recorded in the image descriptor storage means with the image descriptor; And
    Image output means for receiving an image similar to the query image in the image descriptor dissimilarity measuring means from the image database of the database server and outputting the image so that the user can check it.
    Content-based image retrieval system comprising a.
    [2" claim-type="Currently amended] The method of claim 1,
    The image descriptor extraction means,
    A content-based image retrieval system characterized by extracting ART coefficients from an input image based on an ART conversion equation expressed by the following equation.
    (Mathematical formula)

    [3" claim-type="Currently amended] The method of claim 1,
    The query image descriptor extracting means,
    A content-based image retrieval system characterized by extracting ART coefficients from an input query image based on an ART conversion equation expressed by the following equation.
    (Mathematical formula)

    [4" claim-type="Currently amended] The method of claim 2 or 3,
    The ART conversion formula,
    A content-based image retrieval system comprising an ART basis function represented by the following "mathematical formula".
    (Mathematical formula)

    [5" claim-type="Currently amended] The method of claim 4, wherein
    The ART basis function is
    Content-based image retrieval system comprising a declination function Am (θ) defined by an exponential function having a rotational invariance represented by the following equation.
    (Mathematical formula)

    [6" claim-type="Currently amended] The method of claim 4, wherein
    In the ART basis function,
    Radiation function R n (ρ) is a content-based image retrieval system, characterized in that defined by the trigonometric function represented by the following equation.
    (Mathematical formula)

    [7" claim-type="Currently amended] The method of claim 1,
    The image descriptor dissimilarity measuring means,
    Content-based image, characterized in that for measuring the dissimilarity between the image of the query image and the reference image recorded in the image descriptor storage means based on a measurement expression indicating the degree of dissimilarity expressed by the following "mathematical formula" Search system.
    (Mathematical formula)

    [8" claim-type="Currently amended] The method of claim 1,
    The image descriptor,
    The content-based image retrieval system characterized by using the absolute value of the ART coefficient represented by the following formula.
    (Mathematical formula)

    [9" claim-type="Currently amended] The method of claim 1,
    And the image to be searched is input from a remote user web browser.
    [10" claim-type="Currently amended] The method of claim 1,
    The image query means is a content-based image retrieval system, characterized in that for the user to create and input the image to search directly.
    [11" claim-type="Currently amended] The method of claim 1,
    The image query means is a content-based image retrieval system, characterized in that to select and input the image to be searched from the prototype image provided by the web server.
    [12" claim-type="Currently amended] The method of claim 1,
    The image query means is a content-based image retrieval system, characterized in that for inputting an image to search from among the images stored in the recording medium.
    [13" claim-type="Currently amended] The method of claim 1,
    The image query means is a content-based image retrieval system, characterized in that the user reads an existing image, and inputs the image to be changed and searched.
    [14" claim-type="Currently amended] A first step of constructing a database for content-based image retrieval;
    A second step of inputting a query image to be searched using the image input means;
    Extracting an ART image descriptor from the query image input by the image descriptor extracting means to measure dissimilarity between the image stored in the database and the image descriptor, and obtaining a similar image; And
    A fourth step of outputting the image retrieved from the database for the user to check
    Content-based image retrieval method comprising a.
    [15" claim-type="Currently amended] The method of claim 14,
    The first step,
    A fifth step of inputting an image to be stored in a database using the image input means;
    A sixth step of extracting, by the image descriptor extracting means, an image descriptor from an image input to the image input means;
    A seventh step of recording the image descriptor obtained by the image descriptor extracting means into the image descriptor storing means; And
    And an eighth step of classifying the image input by using the image input means by a predetermined criterion and recording the image in the image storage means.
    [16" claim-type="Currently amended] The method of claim 15,
    The sixth step,
    A content-based image retrieval method characterized by extracting an ART image descriptor from an input image based on an ART definition expression expressed by the following mathematical expression.
    (Mathematical formula)

    [17" claim-type="Currently amended] The method according to claim 14 or 15,
    The third step,
    A ninth step of the query image descriptor extracting means extracting an image descriptor from a query image input to an image input unit;
    A tenth step of transferring the image descriptor extracted by the query image descriptor extraction unit to the image descriptor dissimilarity measurement unit;
    The image descriptor dissimilarity measuring unit, which receives the image descriptor of the query image from the query image descriptor extracting unit, measures the dissimilarity with the image descriptor stored in the image descriptor storage unit to determine an image similar to the query image. Step 11; And
    And a twelfth step of acquiring the images corresponding to the image descriptors determined by the image descriptor dissimilarity measuring means from the image storing means, sorting them in the order of similarity, and then transmitting the images to the image output means.
    [18" claim-type="Currently amended] The method of claim 17,
    The ninth step,
    A content-based image retrieval method characterized by extracting an ART image descriptor from an input query image based on an ART definition expression expressed by the following equation.
    (Mathematical formula)

    [19" claim-type="Currently amended] The method of claim 17,
    The eleventh step,
    Content-based image, characterized in that for measuring the dissimilarity between the image of the query image and the reference image recorded in the image descriptor storage means based on a measurement expression indicating the degree of dissimilarity expressed by the following "mathematical formula" Search method.
    (Mathematical formula)

    [20" claim-type="Currently amended] The method of claim 17,
    The twelfth step,
    And the image retrieved by the image descriptor dissimilarity measure means is reused as a prototype image provided by a web server, and the image-based image retrieval method may be modified and re-searched.
    [21" claim-type="Currently amended] The method of claim 14,
    The fourth step,
    And a thirteenth step, wherein the image output means provides the retrieved image as a prototype image when the user reconnects, and transmits the prototype image selected as the input image to the image querying means.
    [22" claim-type="Currently amended] In a content-based image retrieval system equipped with a microprocessor,
    A first function of building a database for content-based image retrieval;
    A second function of inputting a query image to be searched using the image input means;
    A third function of extracting an ART image descriptor from the query image input by the image descriptor extracting unit to measure dissimilarity between the image stored in the database and the image descriptor, and obtaining a similar image; And
    A fourth function of outputting the image retrieved from the database for the user to check
    A computer-readable recording medium having recorded thereon a program for realizing this.
    类似技术:
    公开号 | 公开日 | 专利标题
    KR101887002B1|2018-09-10|Systems and methods for image-feature-based recognition
    US9424277B2|2016-08-23|Methods and apparatus for automated true object-based image analysis and retrieval
    US8868619B2|2014-10-21|System and methods thereof for generation of searchable structures respective of multimedia data content
    Cozzolino et al.2014|Copy-move forgery detection based on patchmatch
    US20150379043A1|2015-12-31|System and Methods for Creation and Use of a Mixed Media Environment
    US20190026268A1|2019-01-24|Associating still images and videos
    US9317534B2|2016-04-19|Combined semantic description and visual attribute search
    Ke et al.2004|Efficient near-duplicate detection and sub-image retrieval
    US8326091B1|2012-12-04|Ranking of images and image labels
    TWI506459B|2015-11-01|Content-based image search
    US8818916B2|2014-08-26|System and method for linking multimedia data elements to web pages
    Li et al.2014|Spatially aggregating spectral descriptors for nonrigid 3D shape retrieval: a comparative survey
    Kim et al.1998|Content-based trademark retrieval system using a visually salient feature
    US7587412B2|2009-09-08|Mixed media reality brokerage network and methods of use
    US8276088B2|2012-09-25|User interface for three-dimensional navigation
    EP2015225B1|2012-10-10|Invisible junction feature recognition for document security or annotation
    US7353224B2|2008-04-01|System and method for efficiently finding near-similar images in massive databases
    DE60033118T2|2007-11-22|System and method for content-based retrieval of images
    EP2054855B1|2012-07-04|Automatic classification of objects within images
    US5845288A|1998-12-01|Automated system for indexing graphical documents having associated text labels
    JP5309155B2|2013-10-09|Interactive concept learning in image retrieval
    US8600989B2|2013-12-03|Method and system for image matching in a mixed media environment
    US8332401B2|2012-12-11|Method and system for position-based image matching in a mixed media environment
    US8107689B2|2012-01-31|Apparatus, method and computer program for processing information
    US7917554B2|2011-03-29|Visibly-perceptible hot spots in documents
    同族专利:
    公开号 | 公开日
    KR100353798B1|2002-09-26|
    US20020010704A1|2002-01-24|
    JP3635368B2|2005-04-06|
    DE60033118T2|2007-11-22|
    DE60033118D1|2007-03-15|
    DK1107136T3|2007-05-29|
    PT1107136E|2007-05-31|
    US20040199531A1|2004-10-07|
    ES2281326T3|2007-10-01|
    EP1107136B1|2007-01-24|
    JP2001209803A|2001-08-03|
    US6754667B2|2004-06-22|
    EP1107136A1|2001-06-13|
    AT352817T|2007-02-15|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    1999-12-01|Application filed by 송문섭, 주식회사 현대큐리텔, 김회율, 김영섬, 주식회사 코난테크놀로지
    1999-12-01|Priority to KR1019990054300A
    2001-07-02|Publication of KR20010053788A
    2002-09-26|Application granted
    2002-09-26|Publication of KR100353798B1
    优先权:
    申请号 | 申请日 | 专利标题
    KR1019990054300A|KR100353798B1|1999-12-01|1999-12-01|Method for extracting shape descriptor of image object and content-based image retrieval system and method using it|
    [返回顶部]