| With the rapid development of Internet, the information is in a state of explosive growth, how to process such kind of large information and make people can easily to find the information needed has increasingly become one of the topic studied, from which the widely used all kinds of search engines now such as Baidu, Google, Bing, etc, have derived.Image retrieval is a part of the search engine. The earlier image retrieval mostly was set up in a manual flagging way, with the results of a low efficiency, subjectivity and so on. As the amount of data increasing, this kind of method obviously can not meet the demand. TBIR (Text-based image retrieval) need the people who upload the image provides necessary tags to describe the image, then the system can store and index the image by the tags.The search engine can find out similar images by the tags which are provided users according to some technique. Currently, most of image retrieval system such as Baidu and Google, use this methods But their image sources are collected by Spider program which can automatically complete to collect the image and extract it’s tags rather than uploaded and tagged manually. This method is deeply depended by the tags provided by users or collected by Spider, ignores the content of image provides by itself, so this method can not ensure the correctness of image search results. Then the CBIR (image search based on image content) was proposed, which received extensive attention. This kind of searching gets the final result by extracting the underlying characteristics of the image and using the similarity among the features of images. Intelligent terminals in recent years, increasingly become more and more matured and with the popularity of mobile network, due to its portability and easy to use, it makes people much more easily to access to the desired information. Most of the current image search is used for PC platforms. Image search of mobile platform, due to its more in line with people’s search habits, will also retrieve the rapid development.This paper tries my best to construct a small commodity image retrieval system, and expand it to the mobile platform. It can be used ultimately to get the relevant information through the mobile platform to photograph or put in the pictures of related product. The work of this paper and the points of innovation are as follows:(1)Build a crawler system for Taobao and Tmall. Such kind of system is just used for collecting wares images and related information, and to a certain degree, acquires the performance of optimization. Transplanting the algorithm of image feature extraction to the Android platform enables it to get feature extraction on the phone, direct return to store links and related goods which are recommended on the basis of the photographs of goods.(2)For about1.2million pictures of feature extraction, this paper adopted the method of LSH (Location Sensitive Hash position Sensitive Hash) combined with tf-idf (term frequency, inverse document frequency). Compared with the linear search which has a good effect, the above method not only makes the result in smaller error range also greatly reduces the time complexity. |
PDF Full Text Request