KR20220097786A - Method for fabric matching using image search - Google Patents

Abstract

The present invention relates to an apparel subsidiary material recommendation learning method using image search, which can classify an apparel subsidiary material image received from a customer into a plurality of groups using a deep learning-based primary classification engine, and uses a separate deep learning-based secondary classification engine optimized for the corresponding group to recommend one or more apparel subsidiary material products to the customer according to the degree of similarity of images among the products in a category within the corresponding group. According to the present invention, the final apparel subsidiary material category is determined through primary classification through a primary classification engine which classifies apparel subsidiary materials into multiple groups according to the characteristics of each group, and through a secondary classification engine for each group which has undergone specialized learning according to the characteristics of the group. In the secondary classification, if a threshold is not exceeded as a result of the classification of the apparel subsidiary material category, it is assumed that group selection in the primary classification has failed, and feedback is given to the next group to attempt the secondary classification. Therefore, accuracy of classification can be increased dramatically.

Description

Fabric matching method using image search {METHOD FOR FABRIC MATCHING USING IMAGE SEARCH}

The present invention relates to a fabric matching method using an image search, and more particularly, to a fabric matching method using an image search for matching one or more fabric products similar to a fabric image uploaded by a user.

The existing online service for auxiliary materials for clothing adopts the form of a general online service shopping mall as it is. It is a method that gives options according to keyword search and various characteristics, and when a suitable product is found, the user selects it.

Clothing subsidiary materials are distinguished from general shopping mall products. The first is that there is no standardization. The second is that the product categories are very diverse. The third is that different names exist for the same product.

Due to the nature of clothing subsidiary materials, the search efficiency is very low in general keyword searches. In order to increase such a search rate, it is necessary to add vast amounts of metadata for various products. However, due to the nature of clothing subsidiary materials, there are things that are created and disappear in various forms, and it is difficult to grasp the market needs for purchase, so it is difficult to build such data proactively, and there is no suitable entity to build it.

The factories and wholesalers that make these products in the market are small, and in order to establish a database for the overall market and systematically prepare to enter the online market, each vendor has to do it themselves, and the data built in this way is also not shared. .

The most suitable way to solve this problem is to find clothing subsidiary materials through image search. Since the main information of clothing subsidiary materials is included in the image, a significant part of the above problems can be solved by securing enough image data.

However, there are some problems.

First, the characteristic of auxiliary materials for clothing is that there are many similar types of auxiliary materials for clothing.

- Button, waffle pen, stopper

- Tape, band, ribbon, zipper

- thread, string

Buttons, waffles, and stoppers are similar in shape in a round shape, tapes, bands, ribbons, and zippers are similar in shape to each other, and threads and strings are similar in shape to each other.

In this case, it is difficult to select an accurate category with only a single image search method, and data that is not suitable for similarity searches may be derived. A method that can accurately derive each characteristic is needed.

Second, since there is no standardized standard for auxiliary materials for clothing, they are created in various forms in the market, and re-learning of auxiliary materials for clothing is required each time. If sufficient learning is not preceded, the accuracy of the search decreases, but a lot of computer resources are required for learning. There is a need for a method that can provide high accuracy while minimizing the resources required for learning.

Third, if the image data input according to the user's search request is reflected in the learning again, the accuracy in learning is continuously increased as the number of users increases. When a user purchases or makes a meaningful action on a product that provides more than an appropriate level of accuracy among the images provided at the request of the user, it can be known that the image is closely related to the product, and supervised learning ( (supervised learning) should be used as data.

The present invention was invented to improve the above problems, and an object of the present invention is to provide a technology that can efficiently improve the search in the existing clothing subsidiary material market.

Using a deep learning-based primary classification engine, first classify which group the clothing subsidiary material image received from the customer belongs to among a plurality of clothing subsidiary material groups, and a deep learning-based secondary classification specialized for each group classification and search The classification engine is used to determine which type of clothing subsidiary material image received from the customer falls under the category of the clothing subsidiary material, and when the clothing subsidiary material category is determined, the secondary classification engine is used An object of the present invention is to provide a method for learning to recommend accessory materials for clothing using an image search for recommending one or more clothing accessory products similar to a received clothing accessory material image.

In order to achieve the above object, the present invention provides an image receiving step of receiving an image of an auxiliary material for clothing uploaded by a customer; a group classification step in which the group classification unit uses a deep learning-based primary classification engine to classify to which group the clothing auxiliary material image received from the customer belongs to among a plurality of clothing auxiliary material groups; Category classification that uses a deep learning-based secondary classification engine to classify which type of clothing auxiliary material image received from a customer belongs to one or more different clothing auxiliary material categories within the group step; and a similar product recommendation unit recommends one or more clothing auxiliary material products similar to the clothing auxiliary material image received by the image receiving unit to the customer based on the degree of image similarity within the classified products of the clothing auxiliary material category using the secondary classification engine. It includes a similar product recommendation step.

The present invention divides the clothing into multiple groups according to the characteristics of the auxiliary material, and the final clothing auxiliary material through the primary classification engine for classifying each group, and the secondary classification engine for each group that has specialized learning according to the characteristics of the group Decide on a category.

1 is a flowchart of a method for recommending and learning clothing subsidiary materials using an image search according to the present invention;

Advantages and features of the present invention, and a method for achieving them will become apparent with reference to the embodiments described below in detail in conjunction with the accompanying drawings.

However, the present invention is not limited to the embodiments disclosed below, but will be implemented in various different forms.

In the present specification, the present embodiment is provided to complete the disclosure of the present invention, and to fully inform those of ordinary skill in the art to which the present invention pertains to the scope of the invention.

And the invention is only defined by the scope of the claims.

Accordingly, in some embodiments, well-known components, well-known operations, and well-known techniques have not been specifically described to avoid obscuring the present invention.

In addition, the same reference numerals refer to the same elements throughout the specification, and the terms used (referred to) in this specification are for the purpose of describing the embodiments, and are not intended to limit the present invention.

In this specification, the singular also includes the plural unless specifically stated otherwise in the phrase, and elements and operations referred to as 'comprising (or having)' do not exclude the presence or addition of one or more other elements and operations. .

Unless otherwise defined, all terms (including technical and scientific terms) used herein may be used with the meaning commonly understood by those of ordinary skill in the art to which the present invention belongs.

In addition, terms defined in a commonly used dictionary are not ideally or excessively interpreted unless defined.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.

1 to 4 , in the method for learning to recommend subsidiary materials for clothing using image search, the image receiving unit 210 receives an image of subsidiary materials for clothing uploaded by the customer ( S110 ). Customers upload images of clothing subsidiary materials they want to search, such as zippers, laces, and buttons, using their computer or smart phone.

The group classification unit 220 classifies which group the clothing subsidiary material image received from the customer belongs to among the plurality of clothing subsidiary material groups by using a deep learning-based primary classification engine (S120). In the present invention, a total of 17 clothing subsidiary material categories were described as an example of classification into 7 groups.

Referring to Table 1, each of the clothing auxiliary material groups includes one or more clothing auxiliary material categories, and the group classification unit 220 provides different clothing auxiliary material categories according to the difficulty and characteristics of the similarity search based on the characteristics of each clothing auxiliary material category. group into multiple groups. It is grouped into A~G groups and the characteristics of categories included in each group are as shown in the example of Table 1 below. That is, although the characteristics are different for each category, the clothing subsidiary material categories with similar tendencies are classified into the same group according to the difficulty and nature of the similarity search.

The group classifying unit 220 classifies to which group the clothing auxiliary material image received from the customer belongs to among a plurality of clothing auxiliary material groups, and stores weight information for each classification group with respect to the image. For the clothing subsidiary material image, weight information for each group A to G is stored.

The category classification unit 230 uses a deep learning-based secondary classification engine to determine which type of clothing auxiliary material image received from the customer is from one or more different clothing auxiliary material categories within the group. Classify whether it belongs (S130).

If the classification result of the category classification unit 230 indicates which clothing subsidiary material category belongs to, it is determined that the category classification is complete, and if the result is less than the threshold value, it is determined that the category classification has failed, and the group classification unit 220 feedback to select a group having the next weight, and the category classification unit 230 performs the category classification step again with respect to the next priority group having the next weight. For example, if the result of group C is less than the threshold, it is determined that the category classification has failed, and the group A, which is the group having the next weight, is selected and the category classification is performed again on the group A.

200... Clothing subsidiary material recommendation learning system [0123]
210...image receiver
220...group division
230...Category Division
240...Similar Product Recommendation Department
250...Product selection result reflection department
260...Subsidiary material similarity calculation department
270... similarity probability value conversion unit
280...Clothing subsidiary materials DB

Claims (1)

an image receiving step of receiving, by an image receiving unit, an image of an auxiliary material for clothing uploaded by a customer;
a group classification step in which the group classification unit uses a deep learning-based primary classification engine to classify to which group the clothing auxiliary material image received from the customer belongs to among a plurality of clothing auxiliary material groups;
Category classification that uses a deep learning-based secondary classification engine to classify which type of clothing auxiliary material image received from a customer belongs to one or more different clothing auxiliary material categories within the group step; and
The similar product recommendation unit recommends one or more clothing subsidiary material products similar to the clothing subsidiary material image received by the image receiver based on the degree of image similarity in the classified products of the clothing auxiliary material category using the secondary classification engine to the customer Including the similar product recommendation step,
The similar product recommendation unit can indicate the characteristics of the clothing subsidiary material category product by using a secondary classification engine for each group that performs specialized learning according to the characteristics of the group to which the category belongs within the product of the clothing subsidiary material category classified by the category classification unit. Based on the second characteristic point in
The secondary classification engine of the similar product recommendation unit compares the pattern of the second characteristic point extracted from the received clothing subsidiary material image within the specific category searched by the category classification unit with the pattern of products belonging to the corresponding category, and the image has a high degree of similarity. Searches for one or more clothing subsidiary material products in the clothing subsidiary material DB and recommends them to customers, and the product selection result reflection unit performs secondary classification when the customer purchases a product with accuracy within a certain error range from the list of products recommended by the similar product recommendation unit Further comprising a product selection result reflection step of increasing the prediction accuracy of the engine and reducing the prediction accuracy of the secondary classification engine when a product having an accuracy outside a certain error range is purchased,
The product selection result reflection unit records the category of the purchased product in the prediction accuracy of the secondary classification engine when the customer purchases another product not recommended by the similar product recommendation unit, and records the purchased product category in the secondary classification engine of the group to which the specific category belongs. A method of learning to recommend subsidiary materials for clothing using an image search, characterized in that it reports to a manager when other category purchase data is accumulated.

Images