Research And Application Of Clothing Style Based On Deep Learning

With the rapid development of social and economic development,residents' consumption level is increasing,the competition of major clothing brands is becoming more and more fierce,and people's requirements for clothing are getting higher and higher.When consumers buy clothing,they pay attention to whether it fits well and have higher requirements for the style of clothing and the style displayed.Studying the fashion styles shown in the 'Big Four' fashion weeks and modeling and quantifying the fashion styles has become a hot research topic today,which is of great significance to the clothing industry.In the existing research on clothing style,most researchers limit the study of clothing style to the global characteristics of clothing.At the same time,clothing style as a subjective judgment,and the relationship between clothing attributes and style is often ignored.This paper relies on the background of the horizontal project of the group's clothing fashion design information platform.In this paper,we extract clothing fashion attributes from clothing images and model clothing fashion styles and summarize the fashion styles of clothing and their applications.The main work of this paper is shown as follows:(1)We built a fashion runway dataset,VogueStyle.Firstly,we extracted highfrequency clothing colors by performing foreground extraction,transforming color space,and high-frequency color clustering on the clothing runway image garments of15 fashion brands at the 'Big Four' fashion weeks in the past five years.Then the clothing styling labels were extracted using the migration learning-based Resnet-50 convolutional neural network,which provides a data source for clothing fashion style modeling.(2)We used a topic model based on word embeddings to model the popular clothing style.Modeling the popular clothing style using the topic model can solve the semantic gap between the clothing style and the popular clothing elements.We proposed a topic model of clothing style combined with deep learning.Furthermore,the clothing styles in the dataset of this paper are classified according to expert knowledge.This paper evaluates the model of this paper by normalizing the mutual information using the external indexes of cluster evaluation,according to which the evaluation shows that the styles summarized by the clothing style theme model combined with deep learning proposed in this paper are closer to the realistic styles compared with the traditional theme models.Finally,the model proposed in this paper is applied to summarize the popular clothing styles over the years and investigate the similarity of clothing images.(3)We designed and implemented a fashion-style display system.The system is based on Spring-Boot architecture,using the MySQL database and Redis database to store data,realizing the functions of user login and registration,clothing style retrieval,clothing trend view and clothing collection,and visualizing a series of data displays.We extracted the high-frequency colors of the clothing by foreground extraction,transforming the color space and high-frequency color clustering.Then we used the migration learning-based Resnet-50 convolutional neural network to extract the clothing style labels.According to the experimental evaluation,the average accuracy of the model was 0.811,and the average precision was 0.880,which was suitable for the clothing style labels.Then,a clothing style topic model combined with deep learning is proposed.Moreover,according to the expert knowledge,the clothing in the dataset of this paper is classified into eight categories Casual,Preppy,Bohemian,Punk,Classic,Streetwear,Sports,and Grunge.This paper evaluates the model of this paper by normalizing the mutual information using the external index of cluster evaluation,.Finally,we build a clothing style display system according to the model proposed in this paper.