Embedded Image Process And Mobile Network Applications With Smartphones

Nowadays, smartphone consists of three functional modules:basic image capturing module, front-end/local processing module and wireless communica-tion module. It becomes the main physical testbed for embedded image process-ing and mobile network applications. Smartphone releases the communication bandwidth and computation overhead by processing the images locally. Tak-ing advantage of the interface for wireless communication enables smartphone’s flexibility and configurability, which motivates the development of embedded im-age processing and mobile network applications based on smartphone. However, since smartphone has limits in resources, image capturing has several disadvan-tages, such as:low resolution, great pollution by noise and easy distraction from environment. Thus all these issues bring great technical challenges to embedded image processing. There are several key technical problems, illustrated as follows: first of all, how to extract accurate information from the polluted data with the limits of sensors’performance is a serious problem. Secondly, smartphone’s abil-ity of computation and the bandwidth of mobile network are all limited. How to efficiently allocate the tasks of image processing between the local and re-mote ends becomes a great challenge if the energy consumption, computation overhead, real-time performance are all considered. Finally, great difficulties of mobile network applications also lie in how to process data and compute big data with large database. This thesis will focus on visible light communication and mobile visual search to demonstrate the challenges in embedded image processing based on smartphone and provide effective solutions.1.Algorithms for image denoising based on feedback informationThere exists serious reflection noise in the images captured by the LCD because of the face-to-face position between the smartphone and LCD in the communication mode. The noise will influence the accurate decoding of color barcode and normal communication. According to the experimental tests, since the images of black screen could effectively measure the reflection noise during the process of communication, this thesis would calibrate the actual influences caused by the reflection noise by inserting black-screen images in the data flows sent based on feedback information. In this way, pollution noise in the images received could be eliminated by using black-screen images.2.Design of fingerprint and specific feature and strategy for fast searching algorithmThere are several challenges in the mobile cloth search system:first of all, there exists distortion when the query is captured since cloth is ductile. Secondly, The query image contains background noise, which will influence the system’s searching performance. Finally, high dimensional features are not suitable for real-time visual searching with the large database. Taking advantage of three-point image capturing model could help to capture the target effectively without high computation cost. This thesis would conduct researches on capturing model and interaction, which would increase the accuracy and real-time performance of mobile image searching. At the same time, this thesis design cloth fingerprint and three-point distance feature to increase accuracy. Finally, this thesis puts forward fast searching algorithm to improve the accuracy of search system with high real-time performance.