Application Of Machine Learning Image Retrieval Technology In Clothing Image Search

Clothing is a category with huge sales in e-commerce websites such as Taobao and Jing Dong.Clothing image retrieval technology can recommend clothing of the same style and characteristics based on the interests of users,which can enhance user experience and provide better services.The text-based image retrieval method requires manual annotation,which cannot accurately describe the characteristics of clothing,and customers cannot accurately describe their needs.The content-based image retrieval method can automatically extract image features,find clothing pictures with similar features in the retrieval dataset,without manual annotation,and the retrieval results are more suitable for users’ needs.Therefore,it has become the mainstream research direction of image retrieval.This article has two innovations as follows:A semi-supervised hash clothing image retrieval model based on Generative Adversarial Network is proposed.Use Generative Adversarial Network to build models.First train the Generative Adversarial Network,then initialize the parameters of the hash network encoder through the discriminator.Use the generator to generate more samples for the training set to achieve data enhancement,and modify the model to be suitable for semi-supervised learning.Experiments show that this method can make use of unlabeled data and improve the performance of the image retrieval network.A clothing image retrieval method based on the positioning of human joint points is proposed.In this paper,OpenPose is used to extract the joint points of the human body in the character area in the clothing image,and the corresponding relationship between multiple human joint points and the clothing area is used to connect the coordinates of the key points of the human body to indirectly obtain the positions of the collar and sleeves of the clothing.RoI pooling is used to obtain the shallow convolution features of the clothing component area,connect the deep convolution features,and merge the convolution features of different levels to obtain the visual and semantic information of the image for clothing image retrieval.Experiments show that the feature fusion of the region of interest can retain more detailed information and improve the retrieval accuracy.