Research On Clothing Image Retrieval Technology Based On Joint Segmentation And Feature Matching

With the development of Internet,shopping in the Net are becoming an important way to shop.Clothing shops occupies a large proportion of online sales platform.Fashion clothing retrieval devotes to improve the user experience of online shopping,the content-based image retrieval becomes a novel online clothing retrieval way.In recent years,the content-based image retrieval has been an important clothing retrieval way,but in clothing domain,it can not retrieve clothing accurately because of the variety of human body posture and the variable of popular elements.What's more,the accuracy of clothing retrieval has an influence on clothing sales volume.So we present a clothing retrieval method based on co-segmentation and feature matching.To resolve the problem of unsatisfied accuracy of the clothing segmentation,a novel segmentation algorithm is proposed based on HOG(Histogram of Oriented Gradients)features and E-SVM(Exemplar Support Vector Machine)classifiers.The co-segmentation algorithm can be divided into three steps:superpixel grouping,E-SVM classifier training,segmentation propagation,which uses an auxiliary dataset.Firstly,user input image is segmented in superpixel,which makes the images in auxiliary dataset into regions.Secondly,some regions are selected to a positive segmentation,which uses HOG information to train E-SVM classifiers.Finally,the cloth is segmented in user image by E-SVM classifiers and segmentation propagation.To resolve the problem of unsatisfied accuracy of the clothing retrieval,a novel clothing retrieval method was presented based on co-segmentation and feature matching.Firstly,the query image was segmented with an auxiliary dataset.Then,the color feature and the bundled feature of clothing regions were extracted after co-segmentation.We calculate the similarities of the features between the query image and the ones in a retrieval dataset and matched them.At last,tho retrieval results were output in descending order accnording to the similarity.