Clothing Retrieval Platform Based On Semantic Information

Online shopping have been favored due to its convenience since it appeared,Clothing is one of the highest online sales of goods,In order to promote the development of online sales of clothing,quickly and accurately search to get the right clothes become the key.In this paper,we propose a semantic based clothing retrieval platform,linking the low-level physical features with high-level semantic features to solve the problems in the process of image retrieval.This paper contains the following items:1? The application of web crawler technology to establish the initial data set.Using web crawler technology to obtain information about the different types of clothing in the major clothing website to build the initial data set,such as thumbnails,prices,purchase volume,etc.The combination of image retrieval and web crawler can deal with the large scale image data processing and improve the performance of online search.2?Using semantic classification method for image classification.The property is robust to the change of garment Comparing with the low level feature,Different visual attributes extracted from a large number of different training data will be used as high-level semantic representation of the image,The physical characteristics of the clothing image are extracted and summarized,and connect to the high level semantic labels,A semantic model based on least squares probabilistic classification algorithm is builted as a standard for classification of images,it is a support for image similarity measurement.3 ? Using of inverted index technology to locate the similar images in the candidates,it will help to reduce the number of similarity measures,and improve the retrieval accuracy,meanwhile,reduce the retrieval timeThe process of apparel retrieval based on the semantic information is as followed:In order to provide data support for retrieval,we first using crawler technology to grabthe site data and construct the initial data set,then extract local description operator for each of the images in data set,the features normalized based on k-means clustering,and representing an image together with color features.The classification algorithm is then used to classify the feature vectors of the original data set to get the semantic model.In the retrieval phase,users uploading an local picture,the image captured by user is not standardized in generally,the influence of the factors such as geometric distortion,occlusion,cluttered background and luminosity change will bring great challenge to the costume retrieval,Therefore,it is needed to do image preprocessing,including image segmentation,feature extraction and other image preprocessing,and then extract the feature of the target area of the image.The extracted features are classified by semantic classification model,and the classification probability vector is obtained,The method of calculating the similarity of images is to measure the distance of feature vectors.The inverted index method is used to retrieve the same or similar information from the data set quickly and accurately,Such as clothing pictures,prices,sources,visits,etc.The experimental results show that the system is robust and can provide high quality retrieval results.