News

Google Glass provides a great foundation for developing various Computer Vision algorithms and applications. Our group is now actively working on a number of different problems and has several pending algorithms to be implemented on this exciting platform.

One objective is to perform landmark recognition, object recognition, and facial recognition. These applications can help people in numerous aspects of their lives, ranging from tourist navigation to hardware training. With object detection and augmented reality, we can provide tools to train junior engineers with visual instruction instead of verbal guidance. We believe this could effectively improve the quality of training while reducing its cost.

Google Glass is also a promising venue in the application of information retrieval. In an ideal scenario, Glass would be able to pull up information on whatever the viewer sees. Traveling to a new place would mean automatically recognizing it and displaying pertinent local information, such as nearby restaurants or places of interest. Catching a glimpse of a movie poster would allow instant identification of the movie and an option to play the trailer. There are no limits in application, but important challenges exist throughout the process of recognition, retrieval, curation, and display of such content.

One final computer vision application our group is particularly interested in is visual saliency detection, which tries to detect where humans look in an image or video. Automated visual saliency detectors attempt to extract the regions that humans are interested in and are a fundamental process for many other computer vision applications, such as object detection and image retrieval. While Glass has no tracking of the human eye, it does provide an insight into this problem by capturing the motion of a person’s gaze as it turns to [...]

The USC Media Communications Laboratory, led by Professor C.-C. Jay Kuo and two PhD students – Harshad Kadu and Jian Li, and the AdShare engineering team have co-developed a new technology, called fast audio matching (FAM), to detect clean and/or degraded audio that has the master sound recording mixed with dialogue and sound effects such as in a movie, TV show or commercial.

This technology extracts inherent audio fingerprints from an audio clip that are robust to manipulations/degradations. The FAM system can detect songs of partial durations, say, less than 15 seconds and provide a confidence level on a scale from 1 to N, where N being a perfect match. It can be run on a personal computer running Linux, MySQL, Apache, and PHP now, and will be implemented in mobile devices with a cloud environment in the near future.

The joint R&D team has continuously worked on system performance optimization to meet real-time processing requirements. The team also keeps developing a more advanced technique that handles sounds distorted by other methods such as pitch shifts, white noise, and instrument swaps.

Moving from HD to Ultra HD (UHD) has become a trend in the display industry as UHDTV systems can provide better user experience such as stronger sensations of reality and superior picture quality. In SMPTE 2013 symposium, it is forecasted that the shipments of UHDTV sets are expected to reach four million units by 2017.

Act as one of the leading groups in video processing research, MCL lab is currently one of the pioneers in exploring UHD video processing technology. The lab has setup a UHD display with high performance playback system. The purpose of the system is to provide the playback of 4K UHD (i.e., 3840×2160 resolution) raw video data in real-time: 60 frame/sec at minimum and up to 120 frames/sec. Besides, to address the concern that a lack of native UHD contents will limit the research, a professional UHD acquisition system will also be available in the lab soon. We expect that this UHD system can help us create more research topics and produce more UHD contents for the researchers who are interested in the technology of video processing.

The UHD system in MCL lab:

Acquisition: Sony 4K professional handheld camcorder
Display: 65” Samsung 4K Ultra HD Smart TV
Playback: High performance Dell Precision workstation supporting 4K raw video playback up to 120 FPS

The Google LA PhD Summit 2014 was held on February 14th at Google LA site. Jian Li, Hao Xu, Jia He and Xin Zhang from MCL attended the event. They were invited to several events from Google including a keynote talk “Music Understanding”, presentations about “Large-Scale Machine Learning”, “Language Understanding”, “Chrome Security” and “Vision + Quantum”. Besides, they had a good chance to talk with leading Google PhDs, and got the opportunities to meet with Google engineers and project managers.

During the information session, engineers from Google mainly introduced researching differences between academia and Google. Two key differences are about available resources and the motivation of the research. Firstly, Google has the most powerful computer center, which offers almost infinite computation resources. They can train very complicated models, adopt more training samples, and obtain results almost instantly with the help from Google. This is crucial for current computer vision research. Secondly, as of motivation, researchers have no pressure on the quantity of publication. Instead, the quality of the publication plays more important role when they publish papers. Additionally, Google has no limit on sharing the work to research community, but they prefer to share the practical work instead of theories only. Several research scientists pointed out that they fundamentally traded the community work to coding when they move from academia to Google.

Besides, they have learned more about the ongoing research work in Google. Hartwig Adam, who is the technical lead manager of the Visual Search team in LA office, shared his work there. His team is focusing on developing computer vision algorithms and a scalable computer vision application such as image and video searching, data mining. He also works for Google Goggles and Glass. They [...]

Dr. Hui Yong Kim, a senior researcher of Broadcasting and Telecommunications Media Research Laboratory of Electronics and Telecommunications Research Institute (ETRI) and also a former adjunct professor at University of Science and Technology (UST) in Korea, visited Media Communication Lab since September 2013. He generously spent some of his short time with us to share his research.

Could you give us a brief introduction about your research experience?

I would like to start from my PhD research experience. I worked on visual surveillance project during my PhD years, and the project required object detection and segmentation from videos. By separating background and foreground, I compressed detected objects and background differently to increase quality of objects area. After graduation, I joined a start-up company because I wanted to get a whole picture of how the real devices are developed. There, I mainly developed H.264 real-time codec software for several video communication systems, such as IP video phone and other consumer products. Because I was a manager in multimedia team, I also needed to care other issues like middle-wares, communication protocols, audio codecs, graphical user interfaces, and even mechanical designs. After few years, I decided to move to ETRI to continue my research path. In ETRI, I participated and contributed to developments of several international standards, including MPEG Multimedia Application Formats and High Efficiency Video Coding (HEVC). During the process, I developed more than 70 patent algorithms.

Could you also talk about your research interest?

My research interests include video/image signal processing and compression for realistic video services, such as UHDTV and 3DTV. They are called as “realistic media”, which implies making things real to human. My research goal is how to maximize visual reality to the user efficiently under [...]