Research And Application Of Clothing Image Retrieval Based On Parts Detection And Segmentation

With the increasing popularity of the Internet and the rise of e-commerce for clothing,more and more users are shopping for clothing through e-commerce platforms,and buying clothing online has gradually become one of the important consumption methods for people,so how to efficiently and accurately retrieve users' favorite clothing from the huge amount of clothing products has become a popular research direction.At present,the clothing image retrieval technique mainly relies on the results of feature extraction of the whole image,which cannot focus on the parts of the clothing,and the background of the clothing image is generally complex,resulting in low accuracy of clothing image retrieval.Therefore,this paper proposes a clothing image retrieval method based on parts detection and segmentation.In this paper,the current popular clothing image retrieval methods are investigated and their advantages and disadvantages are thoroughly studied,and the main work accomplished is as follows:Firstly,this paper transforms the i Materialist Fashion clothing instance segmentation dataset i Materialist Fashion and further divides the sleeves into short,medium and long sleeve categories,and trains Mask R-CNN to build a deep learning model for clothing image detection and segmentation based on a larger sample clothing image annotation dataset,so as to achieve the detection and segmentation of clothing images.Secondly,in the clothing image retrieval,this paper firstly uses Mask R-CNN to detect and segment the clothing images,so as to obtain the information of clothing body,collar parts,sleeve categories and pocket locations;then uses VGG16 to extract 512-dimensional features vectors for clothing body and collar parts respectively;based on these information,the similarity between the clothing to be retrieved and the clothing in the database is calculated one by one.The similarity is calculated as the cosine similarity of 512-dimensional features of the clothing body and collar,as well as the weighted sum of the similarity of the sleeves and pockets,where the sleeves are compared with the similarity of their categories and the pockets are compared with the similarity of their positions.The retrieval results are presented to the user based on the descending order of similarity.The experimental results show that the method can focus on the whole clothing as well as the individual parts,so that it can realize the retrieval of clothing styles.It also allows the user to adjust the similarity weights,so as to return the retrieval results that best meet the user's individual needs.Finally,for the current problem of ignoring the similarity of clothing parts in clothing image retrieval,this paper designs and implements a mall system based on the clothing retrieval method of part detection and segmentation,which realizes clothing goods retrieval by using the retrieval method proposed in this paper.The whole system is implemented by using the technology of front and back-end separation,with the client side using the We Chat mini program framework and the server side using the Spring Boot framework,which fully takes into account the cross-platform and horizontal scalability of the system.The functions of the system mainly include login and registration module,product browsing and retrieval module,shopping cart and payment module,personal center management module,etc.