Research On Related Technical Of Clothing Recommendation

With the development and popularization of the e-commerce,more and more people are bound up in online shopping.Among them,clothing is the largest part of online shopping.However,there are too much clothing categories in online shop,so it's an urgent problem to be solved that quickly find out what he need from vast number of clothing products.In response to this problem,clothing recommendation come into being.According to the user-specific needs,recommend the targeted clothing to meet the needs of users can greatly bring convenience to life.At present,there are some problems in user-specific needs clothing recommendation.It includes the following aspects: 1)The user's needs are not clear,some of the user's needs are difficult to describe in a qualitative manner;2)There exists big semantic gap between highlevel user's needs and low-level clothing image features;3)There are too much clothing categories and styles,the texture,material and other attributes are too complex;4)Clothing pair recommendation is one of the most important branches.But the factors that influencing clothing pair are not clear,which are need to further research.According to the main problems encountered in the current clothing recommendation technology,we propose new clothing recommendation methods.The main research contents can be concluded as follows:1.In order to fill the blank that the lack of weather information in clothing dataset,we build a large clothing dataset called Weather-to-Clothing(Wo C)that contains 13,229 images,which annotated 12 weather categories and 13 clothing attributes.This is the first clothing dataset that contains fully weather information annotation at home and abroad.2.Clothing region detection is a key preprocess stage in clothing recommendation.We first use Faster R-CNN combined with fully convolutional neural network in clothing recommendation industry to detect and segment the clothing region,which can eliminate the background influence in clothing recommendation stage.3.Weather information is the most important factor when people selecting clothing.However,there is big semantic gap between weather information and image features.We use clothing attributes as the latent bridge to narrow the semantic gap between low-level image features and high-level weather information.And we propose weather-oriented cross-modal clothing recommendation algorithm.4.Clothing pairing is a common problem in daily life.We propose a method that putting clothing image into style and matching learning space to get style and matching features.And we use a feature fusion method to fuse the style and matching features.We put forward a new clothing pair recommendation framework.Finally,we compare our method with the state-of-the-art clothing recommendation approaches,which verify that our algorithms perform better than other methods.