News

Congratulations to four members of MCLab, Martin Gawecki, Tsung-Jung Liu, Kuan-Hsien Liu, and Xue Wang for passing their Defenses in the last week.

Martin Gawecki’s dissertation title is “SIGNAL PROCESSING APPROACH TO ROBUST JET ENGINE FAULTDETECTION AND DIAGNOSIS”. His Dissertation Committee includes: Jay Kuo (Chair), Keith Jenkins and Aiichiro Nakano (Outside Member).The Committee was very pleased with his work in applying signal processing techniques to the solution of real-world problems. It demands a good understanding of the whole problem and the access to some valuable data. This bridging effort has been highly appreciated.

Tsung-Jung Liu’s thesis title is “A LEARNING-BASED APPROACH TO IMAGE QUALITY ASSESSMENT”. His dissertation committee members include: Jay Kuo (Chair), Panayiotis Georgiou and Aiichiro Nakano (Outside Member). They were amazed by the quality of his work and an impressive publication list as a result of his research.

Kuan-Hsien Liu’s thesis title is “FACIAL AGE GROUPING AND ESTIMATION VIA ENSEMBLE LEARNING”. His dissertation committee members include: Jay Kuo (Chair), Panayiotis Georgiou, Suya Yu, and Aiichiro Nakano (Outside Member). The Committee was impressed with the novelty of his proposed face-image-based age estimation methods as well as their excellent performance as compared with other state-of-the-art algorithms. The Committee also encouraged him to submit the high quality work to top-ranked journal/conference.

Xue Wang’s dissertation title is “Machine Learning Based Techniques for Biomedical Image/Video Analysis”. Her dissertation committee includes: Jay Kuo (Chair), Sandy Sawchuk (Co-Chair) and Jesse Yan (Outside Member). The Committee was very satisfied with the excellent presentation and results delivered by Xue.

We all wish them the best for their next step.

The high-definition video broadcasting and streaming services are blooming nowadays.  Consumers can enjoy on-demand video services from Netflix, Hulu or Amazon, and watching high-definition (HD) programs becomes the mainstream for video content consumption. According to the latest report, more than half of US population watches on-line movies or dramas. Specifically, the viewers have increased from 37% in 2010 to 51% in 2013. The watched video programs vary in bit rates and resolutions due to the available bandwidth of their networks. Different sizes of video are transmitted at lower bit rates and up-scaled for display on HDTV (e.g., playing a 720p movie on the 1080p screen). This is common in people’s daily life, yet video quality assessment on HD streaming video has not yet been extensively studied in the past.

To fulfill the need, MCLab has released a new video image database recently, which is called MCL-V. This database captures two typical video distortion types in video services. It contains 12 source video clips and 96 distorted video clips with subjective assessment scores.  The source video clips are selected from a large pool of public-domain HD video sequences with representative and diversified contents. Both distortion types are perceptually adjusted to yield distinguishable distortion levels. An improved pairwise comparison method is adopted for subjective evaluation. Furthermore, several existing image and video quality assessment algorithms are evaluated against MCL-V database.  MCL-V is publicly accessible to facilitate future video quality research of the community.

Download link: https://mcl.usc.edu/mcl-v-database/

Seven MCL members attended the Viterbi PhD hooding ceremony on Thursday, May 15, 2014, from 8:30-11:00 a.m. in the Bovard Auditorium. They were Martin Gawecki, Harshad Kadu, Hyunsuk Ko, Kuan-Hsien Liu, Tsung-Jung Liu, Sanjay Purushotham and Xue Wang. Congratulations to them and their families for their accomplishments in completing their PhD program at USC.

Martin Gawecki received a B.S. degree in Electrical Engineering from the University of California, Riverside (UCR) and an M.S. degree in Electrical Engineering from the University of Southern California (USC). Since 2009, he has been a member in the Media Communications Lab at USC, participating in the fields of statistical signal processing, machine learning, and time series analysis. His dissertation, entitled “A Signal Processing Approach to Robust Jet Engine Fault Detection and Diagnosis,” discusses the algorithms that can be used to advance engine health monitoring (EHM) with respect to data from vibration, acoustic, and classical (gas path) sensors. Done in conjunction with the Pratt-Whitney Institute for Collaborative Engineering (PWICE), his work demonstrates the superiority of vibration over acoustic sensors and showcases the ability of machine learning methods to assess performance of real engines based on fully simulated training data.

Harshad Kadu received his Bachelor’s degree from National Institute of Technology, Nagpur in 2008 and Master’s degree from University of Southern California, Los Angeles in 2011. Since then he has been pursuing his PhD in Electrical Engineering with Media Communications Lab at USC. His research interests include 3D human motion capture data analysis, image and music information processing, computer vision and statistical machine learning. His thesis titled ‘Advanced Techniques for Mocap Data Classification and Text Detection’ presents multi-resolution string representation based temporal and spatial domain techniques for human action recognition. His recent work [...]

The Media Communication Lab (MCLab) at USC has a strong commitment to providing an interdisciplinary environment for student researchers, to nourish diversity of application of electrical engineering. Ph.D. candidate Xue Wang, whose research interest lies in biomedical and information processing, has been working on the validation of an evaluation tool for assessing surgical techniques. Capsulorhexis, which is a significant portion of cataract surgery, is of particular interest. Ongoing work on this problem has collaboration between faculty in the Ming Hsieh Department of Electrical Engineering at USC and the Jules Stein Eye Institute (JSEI) at UCLA since June of 2013.

The research aims at developing a valid set of quantitative and qualitative measures of surgical skills to accelerate the training of residents. Currently, capsulorhexis surgical techniques are assessed through an expert-panel-video-review using evaluation questions from the OSCAR and GRASIS evaluation tools.* However, there is considerable inter-observer variability, and direct quantitative questions were the least reliable. Xue’s work combines knowledge in biomedical image/video processing, machine learning, and computer vision to develop an automated quality assessment system to evaluate and improve surgical techniques, such that it is objective, accurate, and reliable. Human inspection is minimized along with the exploration of surgical video understanding via machine learning.

“It’s exciting to put focused and concerted effort in exploring the unknown and trying to find a solution for challenging problems. When you are working with excellent and talented people, you have a more open outlook on the potential of new cross-disciplinary methods”, Xue said. Xue’s work has been presented at AUPO and ASCRS conferences in January of 2014 and April of 2014, respectively.** Her result shows that video/image analysis provides an objective way to measure proficiency in the capsulorhexis part of cataract [...]

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 [...]