Research On Sister Outfit Image Retrieval Technology Based On Deep Learning

The development of Internet big data has promoted the advancement of the fashion apparel e-commerce industry,and the fashion apparel that helps users find their favorite on the e-commerce platform has attracted more and more attention.Sister outfit,as a type of clothing,are associated with the theme of friendship through clothing styles,fabrics and colors,and have a large market among young people.In the process of studying the image retrieval technology of sister costumes,this paper combines the traditional methods and the methods of expressing the characteristics of clothing images in deep learning.The main work contents are as follows:1.In this paper,two kinds of image feature representation methods,traditional features and deep learning features,are first studied and introduced,and their advantages and disadvantages are summarized.In the aspect of traditional features,circular LBP features and color histogram features,which can represent clothing fabric and color,are emphasized.In the aspect of deep learning,convolution neural network is mainly introduced;2.In order to ensure that the extracted color histogram features can correctly express the color distribution characteristics of clothing,without the interference of background color,the YOLO-v3 target detection algorithm is adopted to detect the clothing part of the image,and the clothing image part is intercepted,and the data preprocessing operation is effectively carried out;3.From DeepFashion data set costume-to-shop subset for the label image filtering,choose the details for sister style features images of the study,the clothing image with type including length,collar,sleeve length,design and so on four types of 19 kinds of attribute tags using Inception-v4 network for tabbed classifier training,and then use the network to extract the characteristics of the clothing,in order to improve the sisters with the retrieval accuracy;4.Combining the traditional features of clothing image with the features of convolutional neural network,and in order to improve the retrieval speed,the k-means clustering algorithm is used to cluster the clothing features,and then the similarity between the features in the class is measured by Euclidean distance to obtain the final Top 10 clothing retrieval image;5.The above clothing retrieval algorithm is applied to sister clothing retrieval,and a sister clothing data set is created to verify the validity of the algorithm.