A Android-Based Large-Scale Clothing Image Retrieval System

Mobile E-commerce has recently become an emerging field due to the unprecedented development of smart phones and applications along with the increasing popularity of online shopping. How to quickly find the goods that the customer really need from the vast amounts of commodities has become an urgent problem for the Mobile E-commerce. The traditional commodity search system usually provide text-based retrieval services for customers to find approximate product. But this retrieval method does not take the important image information in account, while the product image contains a lot of information, the buyers’ description of the goods is not always accurate. Therefore, Many e-commerce companies have put in the study of search using image visual features, this retrieval method has been shown to be more natural and intuitive. But the mobile data traffic and low uplink speeds bring many restrictions during the image transmission. so the size of the data sent over the network needs to be as small as possible to reduce latency and improve user experience, an alternate approach is to extract feature descriptors on the phone, compress the descriptors and transmit them over the network. In order to solve these issues, this thesis uses the Android NDK technology and digital image processing technology to build a mobile image retrieval system based on Android, which solves the aforementioned problems reasonably.The main research work and contributions are as follows:Firstly, a feature extraction model based on mobile terminals is proposed to achieve global and local feature descriptors. Due to the pictures taken by mobile phones is too large, but if we use algorithm to compress these pictures, their information will lost. With the upgrading of mobile phone hardware performance, mobile phones have evolved into powerful image and video processing device. The mobile client processes the query image, extracts features such as color, texture and so on, then transmits feature data to the server. The image-retrieval algorithms run on the server using the feature data as query. Using this model it reduces network latency and improve retrieval efficiency.Secondly, a mobile product image search system based on Android is implemented. In front of the mobile client, we use interactive image segmentation algorithm to extract the region of interest(ROI), their color, texture and local feature was extracted, On the other hand, the Euclidean distance and BoW (Bag of Words) are used to match color, texture and SIFT features respectively. What’s more, if users know some of the characteristics of goods, the image search system allows users to provide key words as an auxiliary to help optimize search results and improve the user experience.