The Research On Related Technology Of Clothing Image Retrieval

Nowadays,with the popularity of internet shopping,various categories of goods,e.g.books,cosmetics,electronic equipment,clothing items can be retrieved by their pictures directly on many online shopping websites.Among them,online clothing retrieval has gained much more attention in recent years.However,it is also the most difficult category to retrieval.There are two main reasons: a)the various attributes of clothing,e.g.complex texture and non-rigid deformation;b)the interference of external environment.These factors also make it difficult to recognize fine-grained clothing styles,such as the color,texture,material and so on.Accordingly,we mainly focus on the problem of cross-scenario clothing retrieval and fine-grained clothing style recognition.Recently,cross-scenario clothing retrieval and fine-grained clothing style recognition have been a hot spot and various kinds of method have been proposed.In this paper,we propose to segment the integrated clothing items to retrieval similar products and extract the regional body patches to recognize the clothing style.Besides,domain-adaptive dictionary learning methods are proposed to settle the problem caused by cross scenarios.The main characteristics of the method are show as follows.a)We analyze the existing clothing retrieval and style recognition algorithms comprehensively and find the advantages and disadvantages of this methods.Aiming at the defects of existing methods,we attempt to give the detailed solutions and elaborate the related fields of clothing image retrieval and recognition.b)A new cross-scenario clothing retrieval framework is proposed.First,we presented a hierarchical superpixel fusion strategy according to the visual similarity.Then the clothing items are segmented completely to retrieve the similar clothing images.c)We propose a new method for cross-scenario clothing style recognition by adopting domain-adaptive dictionary learning to increase the adaptability to the scenario.We treat daily clothing images as target domain,online product clothing images as source domain and optimize the target domain through dictionary learning from source domain.d)We collect Product Clothing dataset(PC)and Daily Clothing dataset(DC)which have also been perfectly annotated.Obviously,large discrepancies exist between these two scenarios.Leveraging domain-specific knowledge,we manually identified 55 attributes pertaining to clothing style.In this way,we obtain fine-grained attributes of the clothing items and use the attributes matching score to re-rank the retrieval results further.The experiment results show that our method outperforms the state-of-the-art approaches.