KR20190018195A - Image Concept Determination System and Its Determination Method - Google Patents

Abstract

The present invention relates to an image concept deciding system capable of extracting the most emphasized image concept and a deciding method thereof. According to the present invention, the image concept deciding system comprises: a cooperative filtering module including a similarity weight calculation module, a similar image selection module, an image relation prediction module, a predicted relation evaluation module; and an integrated filtering module including a content-based filtering module and an image-based filtering module.

Description

Image Concept Determination System and Determination Method

More particularly, the present invention relates to an image concept determination system and method for extracting a most emphasized image concept when a plurality of people or goods are crowded, To techniques for individually calculating weights or extracting quantitative image concepts based on statistics.

An image concept is a specific idea that is associated with being emphasized in a specific image. Because it is pursued with unique personality and characteristics in product design and image rendering, quantitative technology is needed to determine, analyze and evaluate image concept in advance.

Accordingly, the present invention proposes a technique of extracting mainstream images from a plurality of groups composed of various images such as objects or persons, and fixing them as emphasized images.

Prior arts similar to the present invention include 1) Korean Patent Laid-Open Publication No. 10-2014-0056085 'Apparel design development simulation system and method for developing clothing design using the same. In the above-described prior art, the design inspection unit extracts image data from a database and outputs the extracted image data; S2 of generating clothing design concept data by image mapping the image data output from the design inspiration unit in correspondence with predetermined design concept data; S3 receiving sketched clothing design sketch data according to the clothing design concept data generated by the first mapping unit for the design sketch virtual model; Step s4 of receiving the design data of the clothing design diagrammed with the clothing design sketch data as input by the diagramming unit; S5 combining the data of at least one of the color data, the pattern data, and the textile data with the pattern-added clothing design diagram data to generate clothing design patterning data; And step S6 of applying the garment design patterning data to the virtual model by the second mapping section and solidifying the data.

Another similar prior art is 2) Korean Patent Registration No. 10-1617828 entitled " Quantitative evaluation network-based image concept information extraction system ". In the above-described prior arts, a plurality of client terminals operated by an image concept evaluator, each client terminal outputs a selection item content for concept extraction for detecting a concept of image content and outputs the selected selection item content for concept extraction A client terminal unit for receiving selection information of an image concept evaluator for the client terminal unit; The client terminal unit is interlocked with each client terminal constituting the client terminal unit. The image concept information contents, the selection item contents for concept extraction and the concept extraction algorithm are set and managed, the selection contents for concept extraction are delivered to the client terminal, And an image concept extraction unit for extracting the concept of the image content by a concept extraction algorithm that receives the evaluator selection information.

However, the conventional art has a problem that it is difficult to extract the most emphasized image concept when a plurality of people or goods are crowded. In addition, there is a problem that it is impossible to calculate weight individually when extracting a specific image from a screen of plural configurations or to quantitatively extract an image concept based on statistics, and no technical means for solving the problem has been proposed.

Korean Patent Laid-Open Publication No. 10-2014-0056085 (2014.05.09.) 'Apparel design development simulation system and method of developing clothing design using it' Korean Patent Laid-Open Publication No. 10-1617828 (May 20, 2016) 'Image concept information extraction system based on quantitative evaluation network'

The present invention aims to satisfy the technical needs required from the background of the above-mentioned invention. In particular, it is an object of the present invention to provide a technique for establishing a single emphasized image concept in a plurality of persons or objects of a variety in which diversity exists.

In order to quantitatively estimate the degree of emphasis in a crowded situation, a plurality of images extracted from the respective clusters are classified into a plurality of images There is a further object to provide a technique of calculating a weight for a plurality of clusters and determining an emphasized image concept as an average of values extracted from a plurality of clusters.

The technical objects to be achieved by the present invention are not limited to the above-mentioned problems, and other technical subjects not mentioned can be clearly understood by those skilled in the art from the following description. There will be.

According to an aspect of the present invention, there is provided an image concept determination system including a similarity weight calculation module for calculating a similarity weight between specific image data and similar image data, a similarity calculation module for determining a sample population of image data having high similarity to specific image data, A collaborative filtering module including an image selection module, an image association degree prediction module for predicting a degree of association between the sample group and the image data of the sample group, and a prediction association degree evaluation module for verifying the validity of the predicted association degree; A content-based filtering module for ranking and filtering the text similarity degree of an image concept representing image data, and an integrated filtering module including an image-based filtering module for extracting and filtering pixel characteristics of image data, A method for determining an image concept using a system includes a similarity weight calculation step of calculating a similarity weight among image data, a similar image concept selection step of determining a sample population of highly similar image data, an image concept association prediction A cooperative filtering step in which an image concept prediction association degree evaluation step of checking the validity of the predicted association degree is included; An image concept content-based filtering step of ranking and filtering the text similarity degree of the image concept, and an image concept information-based filtering step of extracting and filtering pixel characteristics of the image data. .

As described above, the present invention has the effect of confirming the emphasized image concept among the clusters of different images. In addition, there is an effect that can be applied to an image making operation in which a specific image must be confirmed from a specific cluster. In addition, the concept of an individual store composed of a plurality of commodities is deduced and determined by a quantitative method in a place where a plurality of shops are concentrated, and thus the system can be used for editing shops or commercial analysis.

It is to be understood that the technical advantages of the present invention are not limited to the technical effects mentioned above and that other technical effects not mentioned can be clearly understood by those skilled in the art from the description of the claims There will be.

1 is a configuration diagram of an image concept determination system according to the present invention;
2 is a block diagram of a cooperative filtering module according to the present invention;
3 is a configuration diagram of an integrated filtering module according to the present invention;
4 is a flow chart of an implementation of cooperative filtering in accordance with the present invention;
Fig. 5 is a flowchart of the implementation of the image concept integrated filtering.

BRIEF DESCRIPTION OF THE DRAWINGS The above and other objects, features and advantages of the present invention will become more apparent from the following detailed description of the present invention when taken in conjunction with the accompanying drawings, It is not. In the following description of the present embodiment, the same components are denoted by the same reference numerals and symbols, and further description thereof will be omitted.

The image concept determination system 10 for extracting an image concept for a plurality of products according to the present invention and determining a representative image concept includes a similarity weight calculation module 110 for calculating a similarity weight between specific image data and similar image data, A similarity image selection module 120 for determining a sample population of image data having high similarity to image data, an image association degree prediction module 130 for estimating a degree of association between the sample population and image data of the sample population, A cooperative filtering module (100) including a prediction association evaluation module (140) for verifying the validity of the association; An integrated filtering module 200 including a content based filtering module 210 for ranking and filtering the text similarity degree of an image concept representing image data, an image based filtering module 220 for extracting and filtering pixel characteristics for image data, ; ≪ / RTI >

The image concept determination method using the image concept determination system 10 includes a similarity weight calculation step S110 for calculating a similarity weight value between specific image data and similar image data, a sample population of image data having high similarity to the specific image data, A similarity image concept selection step S120, an image concept association prediction step S130 for predicting a correlation between the sample population and the image data of the sample population, an image concept prediction association for checking the validity of the predicted association degree A collaborative filtering step (S100) in which the step (S140) of evaluating is included; An image concept content-based filtering step S210 for ranking and filtering the text similarity degree of an image concept representing image data, an image concept including an image concept information-based filtering step S220 for extracting and filtering pixel characteristics for image data An execution flow is formed in the integrated filtering execution step (S200).

The system and method for determining an image concept according to the present invention extracts a specific image at a specific place and displays the specific image as a main emphasized image to conceive an image. The image concept determination system according to the present invention searches for emphasis points or feature points in one or more image content communities and extracts a main image concept corresponding to the emphasis points or feature points as a representative concept of the at least one image content community. The image content in the present invention means image data including characters and objects on-line. In order to extract the above-described emphasis points or feature points, the dictionary information must first be extracted from the image data. In order to extract the dictionary information, area calculation of the image data is required, which is possible by using pixel values of image data.

In one embodiment, in the presence of a particular store having a large number of product groups, it is utilized to find the characteristics of the specific store. For example, there are shops A, B, and C, and stores A total of 35 adidas including 21 flower-patterned clothes, 9 polka-dotted clothing, and 5 non-woven clothes, A total of 70 garments including 25 floral garments, 30 polka dot garments, and 15 uniform garments. In the shop C, a total of 50 points including 27 floral prints, 20 polka dots, Of clothes,

After obtaining the image data of each of the garments stored in the shops A, B, and C, the image data is classified according to the garment pattern, and the image concept name is added to the image data occupying the largest number of the classified image data .

If it is confirmed through the above process that the garment being held by a particular store has the greatest floral pattern, the image of the particular shop is classified as a main emphasized image, and the image concept is given as a romantic name, The concept becomes definite.

In yet another embodiment, when searching for an image concept for a person in a community, a plurality of specific images are extracted from the community into a large classification community, and the plurality of specific images are weighted for the entire image, And the image having the weight is determined as the concept of the community.

For example, there are 25 female celebrity image communities, 7 female celebrities with good eyes, 4 female celebrity with beautiful nose, 2 female celebrity with slim face, 2 female celebrity with beautiful face , And a total of 25 female celebrity images by five other female celebrity images,

Considering that the snow most distributed among the above communities is regarded as a group of good female entertainers, the community is named as good female entertainers and the image concept is confirmed.

And the classification of the image concept is a concept that can be applied to both objects or people. The classification of the objects refers to classifying the objects according to features such as clothing, fashion items, color, and pattern, and the person may be a person who wears the object, a face shape of the person itself, But the invention is not limited thereto.

FIG. 1 is a configuration diagram of an image concept determination system according to the present invention, FIG. 2 is a configuration diagram of a cooperative filtering module according to the present invention, FIG. 3 is a configuration of an integrated filtering module according to the present invention,

The image concept determination system (10) of the present invention includes a similarity weight calculation module (110) for calculating a similarity weight value between specific image data and similar image data, a similarity image selection module A module 120, an image association degree prediction module 130 for estimating a degree of association between the sample group and the image data of the sample group, and a prediction association degree evaluation module 140 for verifying the validity of the predicted association degree A cooperative filtering module 100 included;

An integrated filtering module 200 including a content based filtering module 210 for ranking and filtering the text similarity degree of an image concept representing image data, an image based filtering module 220 for extracting and filtering pixel characteristics for image data, .

The similarity weight calculation module 110 may improve the accuracy of the similarity weight calculation result by selecting one or more of the correlation coefficient value, the vector similarity value, and the basic association degree value, which will be described in detail later.

The content-based filtering module 210 includes a morpheme analysis module 211 for analyzing a morpheme of a text representing an image concept on the basis of a normal noun, a proper noun, and a dependent noun, a feature extracting feature characteristic expression A word extracting module 212, and a profile learning module 213 for providing a concept of image data based on a profile having representation information of an image concept.

The image-based filtering module 220 includes an image feature extraction module 221 for extracting features of similar color histograms of image data, a wavelet transform module 222 for complementing color information of image data, An image clustering module 223 for clustering image data according to the distance range of the color histogram having the maximum value of the color histogram, and an image classification module 224 for classifying the image data into a cluster having the maximum similarity value between the specific image data and the new image data .

The image concept referred to in the present invention is composed of three items of a major classification concept, a middle classification concept and a small classification concept,

The above-mentioned major classification concept is classified into five types: Manish type, Classic type, Casual type, Unique type and Feminine type;

The concept of the large classification of the Manish type is classified into two classes of the classical type and the casual type. The classical concept of the classical type is classified into one classical concept, and the classical concept of the casual type is classified into casual type 1 And the concept of middle classification is classified into two types of classic type and casual type. The concept of the large classification of the feminine type is divided into two classifications of classical type and casual type;

In the Manish type main categories, the sub classifications of the classic type are divided into two sub categories, Manish Classic type and Military Classic type. In the Manish type main category, Casual type sub categories are classified into Manish casual type and Military casual type. Are classified;

The subclass of casual type classified into the above-mentioned classical classifications is classified into three classifications of ModernSec Classic class, SimpleClass class and Yupee Classical class;

In the above casual type main category, the sub classifications of the casual type are classified into five categories: Urban Daily Casual Type, City Casual Type, Jeffy Casual Type, Fashionable Casual Type and Easy Casual Type;

The subclass of the classic type is classified into one of the above-mentioned unique type, and the subclass concept is classified into one of the orientalism class. The subclass of the casual type is classified into the art casual, vintage casual, avant-garde casual, ethnic casual, hip-hop punk casual A small subcategory concept is categorized;

In the above-mentioned feminine type main category, the sub classifications of the classic type are classified into six categories of Romantic Classic, Feminine Classic, Gullish Classic, Elegance Classic, Luxury Classic and Sexy Classic, and the casual type sub classifications of Feminine type are classified into Feminine It is preferable to be classified into three categories of casual, casual, casual and sexy casual.

FIG. 4 is a flow chart of the implementation of the cooperative filtering according to the present invention, FIG. 5 is a flow chart of the implementation of the image concept integration filtering,

The image concept determination method according to the present invention includes a similarity weight calculation step (S110) of calculating a similarity weight value between specific image data and similar image data, a similar image concept selection step of determining a sample population of image data having high similarity with specific image data (S140) for estimating a degree of association between the sample group and the image data of the sample group, an image concept prediction association evaluation step (S140) for checking the validity of the predicted association, (S100);

An image concept content-based filtering step S210 for ranking and filtering the text similarity degree of an image concept representing image data, an image concept including an image concept information-based filtering step S220 for extracting and filtering pixel characteristics for image data An execution flow is formed in the integrated filtering execution step (S200).

In the similarity weight calculation step (S110), the weight is calculated by a correlation coefficient calculation step (S111) of calculating a similarity weight using a Pearson correlation coefficient method or a Spearman correlation coefficient method, a similarity weighting value (S113) of calculating the similarity weight by further considering the image data for which the degree of association is not evaluated with the result value of the correlation coefficient calculating step (S11) The average value of similarity weights calculated using the average value of similarity weights,

The similarity weight value between the specific image data and the similar image data only with any one of the similarity weight values calculated in the correlation coefficient calculation step S111, the vector similarity calculation step S112, and the basic association degree evaluation step S113, .

The Pearson correlation coefficient method is calculated by the following equation (1)

...수식(1)

... (1)

here,

는 상점

가 보유한 상품

의 이미지컨셉에 대하여 평가한 개별 연관도,

Shop

Goods owned by

The individual concept of the image,

는 상점

가 보유한 상품

의 이미지컨셉에 대하여 평가한 개별 연관도,

Shop

Goods owned by

The individual concept of the image,

는 상점

가 보유한 상품의 이미지컨셉에 대한 연관도 평균값,

Shop

The average value of association with the image concept of the product possessed by the user,

는 상점

가 보유한 상품의 이미지컨셉에 대한 연관도 평균값을 의미한다.

Shop

The average value of the associations with respect to the image concept of the product possessed by the user.

The Spearman correlation coefficient method is calculated by Equation (2)

...수식(2)

... (2)

here,

는 상점

가 보유한 상품

의 이미지컨셉에 대한 연관도 순위,

Shop

Goods owned by

Relational rankings for image concepts,

는 상점

의 연관도를 평가한 상품의 이미지컨셉에 대한 중간 순위,

Shop

The intermediate rank of the image concept of the product which evaluated the degree of association of the product,

는 상점

가 보유한 상품

의 이미지컨셉에 대한 연관도 순위,

Shop

Goods owned by

Relational rankings for image concepts,

는 상점

의 연관도를 평가한 상품의 이미지컨셉에 대한 중간 순위이다.

Shop

Is an intermediate ranking for the image concept of the product that evaluated the degree of association.

The vector similarity method is calculated by Equation (3)

...수식(3)

... (3)

here,

는 상점

가 보유한 상품의 이미지컨셉에 대하여 연관도를 평가한 상품의 집합이다.

Shop

Is a set of products that evaluate the degree of association with the image concept of a product possessed by the user.

The basic association degree evaluation method is calculated by Equation (4)

... (4)

here,

는 상점

와 상점

모두에서 이미지컨셉의 연관도가 평가된 상품,

Shop

And shops

In all, products whose relevance of image concept is evaluated,

은 상점

와 상점

중 어느 한 상점에서 이미지컨셉의 연관도가 평가된 상품의 개수,

Silver store

And shops

The number of products in which the degree of association of the image concept is evaluated in any one of the stores,

는 상점

와 상점

모두에서 이미지컨셉의 연관도가 평가되지 않은 상품의 개수이다.

Shop

And shops

In all, the number of products for which the degree of association of the image concept is not evaluated.

Referring to FIG. 4, the similar image concept selection step S120 includes a similarity calculation step S121 for calculating a similarity between specific image data and other image data, and a similarity calculation step S122 for predicting the image concept using the calculated similarity Further comprising a sample image data number determination step (S122) of determining a sample image data number,

The step of determining the number of sample image data (S122) may include an optimum threshold value setting step (S122a) of determining the number of sample image data to use only image data whose similarity weight value between image data is equal to or greater than a reference value, And a similar image data maximum allowable number setting step (S122b) for setting the maximum allowable number.

The image concept association prediction step S130 predicts the association degree of the new image data in which the degree of association is not calculated based on the degree of association between the image data having the similarity degree, Lt; / RTI >

...수식(5)

(5)

here,

는 상점

가 보유한 상품

의 이미지컨셉에 대해서 연관도를 예측한 값,

Shop

Goods owned by

A value obtained by predicting the association degree with respect to the image concept of the image,

는 상점

가 보유한 상품의 이미지컨셉에 대한 연관도 평균값,

Shop

The average value of association with the image concept of the product possessed by the user,

는 상점

와 상점

의 유사도 가중치,

Shop

And shops

The similarity weight,

은 상점

와 다른 상점들 간의 유사도가 0이 아닌 상점의 수이다.

Silver store

And the number of stores whose similarity between the other stores is not zero.

Referring to FIG. 5, the image concept content-based filtering step S210 includes a target web information collection step S211 for collecting web information of a target web site in order to collect text about an image concept, An entry file pre-processing step (S213) for pre-counting the number of image concept texts and the appearance frequency, an image concept text embedded in the entry file is converted into a database The image concept text database construction step S214 is further included. At this time, the image concept text extraction step S212 includes a web document collection step S212a for collecting a web document including a plurality of image concept texts, an entry file storing the number, content, and URL of the collected web documents An entry file creation step (S212b) is further included.

The image concept information-based filtering step S220 includes a color histogram calculation step S221 for calculating a color histogram of the color information of the image data, a shop for analyzing the correlation between color information of the image data, A shop color histogram correlation analyzing step (S223) for analyzing a correlation between the image data color information correlation (S222), the store correlation and the color histogram, An automatic classification step S224 is further included.

While the present invention has been described in connection with what is presently considered to be practical exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, but, on the contrary, will be. Accordingly, the true scope of the present invention should be determined only by the appended claims.

10: Image concept confirmation system
100: Collaborative filtering module
110: Similarity weight calculation module
120: Similar image selection module
130: image association prediction module
140: Predictive relevance evaluation module
200: Integrated Filtering Module
210: Content Based Filtering Module
211: Morpheme module
212: feature word extraction module
213: profile learning module
220: Image-based filtering module
221: Image feature extraction module
222: Hars wavelet transform module
223: Image cluster module
224: Image classification module

Claims (10)

An image concept determination system for extracting an image concept for a plurality of products to determine a representative image concept.
A similarity weight calculation module 110 for calculating a similarity weight value between specific image data and similar image data,
A similar image selection module 120 for determining a sample group of image data having high similarity to specific image data,
An image association degree prediction module 130 for estimating a degree of association between the sample group and image data of the sample group,
A cooperative filtering module (100) including a prediction association evaluation module (140) for verifying the validity of the predicted association;
A content based filtering module 210 for ranking and filtering the text similarity degree of the image concept representing the image data,
And an integrated filtering module (200) including an image-based filtering module (220) for extracting and filtering pixel features for image data. The method according to claim 1,
The similarity weight calculation module 110 selects one or more of the correlation coefficient value, the vector similarity value, the basic association degree value, and the inverse user frequency value to improve the accuracy of the similarity weight calculation result . The method according to claim 1,
The content-based filtering module 210,
A morpheme analysis module 211 for analyzing a morpheme of a text representing an image concept based on a normal noun, a proper noun, and a dependent noun;
A feature word extraction module 212 for extracting a characteristic expression of text representing an image concept,
And a profile learning module (213) for providing a concept of image data based on a profile in which presentation information of an image concept is included. The method according to claim 1,
The image-based filtering module (220)
An image feature extraction module 221 for extracting characteristics of similar color histograms of image data,
A Har wavel transform module 222 for supplementing color information of image data,
An image clustering module 223 for clustering image data according to a distance range of a color histogram having the same length interval,
And an image classification module (224) for classifying the image data into a cluster having a maximum similarity value between the specific image data and the new image data. A method for determining an image concept by extracting an image concept for a plurality of products and establishing a representative image concept,
A similarity weight calculation step (S110) of calculating a similarity weight value between specific image data and similar image data,
A similar image concept selection step (S120) of determining a sample group of image data having high similarity to specific image data,
An image concept association prediction step (S 130) of predicting a correlation between the sample group and image data of the sample group,
A collaborative filtering step (S100) including an image concept prediction association evaluation step (S140) for checking the effectiveness of the predicted association degree;
An image concept content-based filtering step (S210) for ranking and filtering the text similarity degree of an image concept representing image data,
(S200) including an image concept information-based filtering step (S220) for extracting and filtering pixel characteristics of image data. 6. The method of claim 5,
In the similarity weight calculation step S110,
A correlation coefficient calculating step (S111) of calculating a similarity weight using a Pearson correlation coefficient method or a Spearman correlation coefficient method,
A vector similarity calculating step (S112) of expressing specific image data and similar image data in a vector space to calculate a similarity weight,
(S113) for calculating the similarity weight by further considering the image data for which the degree of association is not evaluated with respect to the result value of the correlation coefficient calculation step (S11) are sequentially performed, and the average value of the calculated similarity weight values Use,
The similarity weight value between the specific image data and the similar image data only with any one of the similarity weight values calculated in the correlation coefficient calculation step S111, the vector similarity calculation step S112, and the basic association degree evaluation step S113, Wherein the image is used as an image. 6. The method of claim 5,
The similar image concept selection step (S120)
A similarity calculation step S121 for calculating the similarity between the specific image data and the other image data,
Further comprising a sample image data number determination step (S122) of determining a sample image data number to predict an image concept using the calculated similarity,
The sample image data number determination step (S122)
An optimum threshold value setting step (S122a) of determining the number of sample image data to use only image data whose similarity weight value between image data is equal to or greater than a reference value;
And a similar image data maximum allowable number setting step (S122b) for setting a maximum allowable number of image data similar to specific image data. 6. The method of claim 5,
The image concept content-based filtering step (S210)
A target web information collecting step (S211) of collecting web information of a target web site in order to collect text about an image concept,
An image concept text extracting step (S212) of extracting texts for a plurality of image concepts by the web robot agent,
An entry file preprocessing step (S213) for pre-counting the number of image concept texts and appearance frequency,
Further comprising an image concept text database construction step (S214) of converting the image concept text embedded in the entry file into a database. 9. The method of claim 8,
The image concept text extraction step (S212)
A web document collection step S212a for collecting a web document including a plurality of image concept texts,
And an entry file creation step (S212b) of creating an entry file in which the number, content, and URL of the collected web documents are stored. 6. The method of claim 5,
The image concept information-based filtering step (S220)
A color histogram calculating step (S221) of calculating a color histogram of color information of the image data,
A store and image data color information correlation analysis step (S222) for analyzing a correlation between a store-related association and color information of image data,
A store and color histogram correlation analysis step (S223) for analyzing a correlation between the store-related association and the color histogram,
Further comprising an image data automatic classification step (S224) of classifying a plurality of image data by an image concept.

Images