2023, Vol. 8, No. 1. - go to content...

Permanent address of this page - https://kostumologiya.ru/en/03tlkl123.html

Метаданные этой статьи так же доступны на русском языке

Full article in PDF format (file size: 755.8 KB)

For citation:

Smirnov E.E., Kostyleva V.V., Razin I.B., Murtazina A.R. Recognition of the type of clothing and shoes by image. Journal of Clothing Science. 2023; 8(1). Available at: https://kostumologiya.ru/PDF/03TLKL123.pdf (in Russian).

Recognition of the type of clothing and shoes by image

Smirnov Evgeniy Evgenʹevich
Russian State University named A.N. Kosygin (Technology. Design. Art), Moscow, Russia
E-mail: evg7162@mail.ru
RSCI: https://www.elibrary.ru/author_profile.asp?id=1097456

Kostyleva Valentina Vladimirovna
Russian State University named A.N. Kosygin (Technology. Design. Art), Moscow, Russia
E-mail: kostyleva-vv@rguk.ru
RSCI: https://www.elibrary.ru/author_profile.asp?id=353612

Razin Igor Borisovich
Russian State University named A.N. Kosygin (Technology. Design. Art), Moscow, Russia
E-mail: razin-ib@rguk.ru
RSCI: https://www.elibrary.ru/author_profile.asp?id=850439

Murtazina Alʹfiya Rustyamovna
Russian State University named A.N. Kosygin (Technology. Design. Art), Moscow, Russia
E-mail: aly1029@yandex.ru
RSCI: https://www.elibrary.ru/author_profile.asp?id=906389

Abstract. The article proposes a solution to the problem of searching for information about a product in conditions of limited data about the product itself and knowledge about the subject area, in particular, recognition and classification of types of clothing and footwear in the image. This will eliminate the following difficulties: (1) how any person who is not a professional in the clothing and footwear industry can quickly find information on the Internet about a particular product; (2) collecting statistics on used clothes and shoes by processing photos as part of promotions with hashtags on social networks, etc. As part of this study, experiments were carried out with a fully connected neural network architecture in the problem of object class recognition on the example of a dataset of 60,000 images with one color channel in grayscale. The purpose of the research was to assess the acceptability of the architecture of a fully connected neural network for recognizing light industry products. It is shown that for the consumer, when choosing a product, searching for information about it, or buying, an image with its description is preferable. If the work on its fragmentation and the selection of significant features is shifted to the neural network, then this will reduce the time for creating a training sample, but increase the time for training the network itself. A fully connected neural network recognizes patterns of simple images with an accuracy of about 88 %. Testing of various activation functions showed that in this problem the choice of function is not critical. The results were identical, however, if it becomes necessary to build a deeper network, then the choice of the activation function of the hidden layers will have a greater influence on the result. Such solutions are supposed to be developed within the framework of the doctoral dissertation of Smirnov E.E. and use in the educational process the departments «Artistic Modeling, Design and Technology of Leather Products», «Information Technology» of the Russian State University. A.N. Kosygin (Technology. Design. Art) in the form of textbooks.

Keywords: neural network; machine learning; data processing; image recognition; image classification; deep learning; classification; light industry products

Download article in PDF format