Mobile Visual Search Based On Saliency

With the rapid development of mobile devices, including smartphone and laptop, there is a great improvement in the competence of mobile communication and multi-media processing. These devices provide an appropriate platform for mobile visual search. Mobile visual search is a process of taking photos by mobile device and transmitting image information to remote server via wireless network, then retrieving in database on the server side and giving the feedback to the mobile device. In this process, how to decrease retrieval latency is still a challenge to the industry, which is also the main research topic in this paper.Due to the restriction of network bandwidth, it takes a lot time to transmit an image to the server side, leading to bad user experience in traditional mobile visual search. Because of the start-up phase of mobile visual search, there is still no authoritative retrieval procedure. This paper will focus on increasing retrieval accuracy and decreasing retrieval latency, choosing proper image descriptor and optimizing it aiming at mobile visual search. On the server side, bag of words model is adopted, which is also improved by the relevant transmission strategy. The improved model of mobile visual search can be used on smartphone, laptop and wearable device in the future.This paper introduces some background and researching meaning of this topic briefly, and makes a detailed introduction of the key technologies involved in this paper, including image descriptor, saliency feature and so on. Then, the system framework of mobile visual search proposed in this paper is illustrated in detail, along with the main idea and the realization of the core algorithm. In the final, extensive experiments have been done. Comparing to traditional image retrieval method with relevant parameters: the sum of image descriptors, the amount of transmitted data and the image retrieval rate, experiments demonstrate that with the help of the method proposed in this paper, the efficiency of network transmission can be improved by transmitting less image descriptors to the server side, and achieve higher image retrieval rate simultaneously.