Advanced Signal Processing for Mobile Social Media Sharing

点击翻译按钮获取中文摘要

Recently, sharing images/videos online has become increasingly popular, thanks to advances in wireless networks, smart phones, and social networks. This leads to explosive growth in the number of online images/videos. To reduce uploading cost and delay and the server storage cost, social network providers downsample the high-resolution images/videos before uploading, which wastes the capability of the camera. On the other hand, when other users view these images/videos, they frequently want to zoom in to see some details, but the quality of existing zooming algorithms is still not satisfactory, especially for panoramic images. ******Therefore, how to improve the mobile social media sharing quality without increasing the cost to social network users and providers is a challenging and important problem. Fortunately, some new theories have been developed, and various powerful cloud computing platforms have become available, offering potential opportunities to address the challenge.******The long-term objective of our research is to develop algorithms and systems to improve multimedia communications experience. In the short term, we will develop and integrate novel theories in the following three closely related topics in graph, sparse, and redundant signal processing, apply them to mobile social media sharing, and develop a mobile phone app.******1. Graph transform is a new theory to process data defined on graphs such as those from social networks and sensor networks. It has also been used in image/video coding. However, existing graph transforms do not have fast implementations. In this program, we will design fast graph transforms by generalizing the conventional filter bank theory.******2. Compressed sensing (CS) is a new way to sample and reconstruct sparse signals. Currently, the approximate message passing (AMP) method represents the state of the art in CS. Recently we develop a side information-based AMP (SI-AMP) to incorporate the initial estimate of the signal. In this program, we will further improve SI-AMP by exploiting the latest denoising methods. We will also design a general multi-resolution CS scheme, apply SI-AMP to it, and integrate them with the graph transform.******3. Dictionary-based representation is widely used in CS. In this program, we will design a quantized dictionary-based representation to reduce the storage cost, which is suitable for mobile social media sharing, and apply it to our CS schemes.******4. Mobile app development: we will develop a mobile phone app that allows users to upload and view photos via our app and server. The encoding and decoding of images/videos are based on the algorithms developed above, which will provide better quality than existing schemes.******This program will provide solid theoretical training and product development experience to the students involved, so that they can make contributions to Canadian economy in the future.**