Research And Implementation Of Large-scale Commodity Image Retrieval Based On Mobile Crowdsourcing

With the rapid development and popularization of mobile e-commerce,how to quickly find the goods that meet their needs from the massive commodity images has become a new challenge for mobile e-commerce in large data environment.Traditional commodity image retrieval usually uses the following two ways:1)keyword-based commodity image retrieval technology;2)content-based commodity image retrieval technology.Because the above methods have some shortcomings such as the low retrieval accuracy and the high cost of pretreatment,the efficiency of image retrieval is not satisfactory.This paper first attempts to apply the crowdsourcing model to the mass-based product image retrieval based on content.As we all know,the crowdsourcing model is a new model of the Internet,which uses the power of personal choice and group,plays important roles in the group’s knowledge,skills,information and skills to address diversity and complexity issues.Now there are a large number of successful application examples,such as Google translation,Wikipedia,proper way translation,Baidu Encyclopedia and so on.In the mobile wireless network environment,the network bandwidth and traffic restrictions make the image transmission time-consuming,so this paper proposes an interactive active transmission strategy to improve the image transmission efficiency.The main contents and contributions of this paper are as follows:First,a massive commodity image retrieval framework(that is iCrowd)is proposed.In the iCrowd platform,a query result set based on texture features such as a color histogram,and a retrieval image are made into a crowdsourcing task,which are released in the form of crowdsourcing.And then the results of the crowdsourcing task are used for statistical analysis to obtain the result set required by the query user.Through the iCrowd platform for product search,retrieval efficiency and customer satisfaction significantly improved.Second,an interactive active transmission model is established.The method divides the transmission image into a plurality of image blocks,and the user clicks on the region of interest to keep the resolution of the region of interest image unchanged so that the resolution of the non-region of interest image is correspondingly reduced.This can reduce the whole picture of the transmission size to achieve the desired transmission efficiency.Thirdly,a cache strategy based on K-nearest neighbor is proposed.Using the historical query image can directly obtain some result sets,which can reduce the number of crowdsourcing tasks,so that the query efficiency is greatly improved.