CLARK OLSON: Explores Object Recognition Research while on Sabbatical
Dr. Olson's current research project involves matching keypoints (areas of particular interest in images) of the same object in separate images, even when the images were taken from different perspectives (and sometimes different lighting and backgrounds). Along with his research students, he is studying how to use multiple image cues, including shape, color, and texture, to perform improved matching between images. These keypoints can also be used for object recognition to classify objects in an image into categories such as tiger, chair, person, building, and so on. Classification techniques often use a ‘bag of visual words’ strategy, where keypoints are clustered in a high-dimensional descriptor space and classification can be performed using the frequency of each visual word. Dr. Olson says that this strategy has yielded some of the best results to date, but classification performance is still far from perfect.
Over the last few years, Dr. Olson has transitioned to focusing on the related problems of content-based image retrieval and image classification. His sabbatical afforded the opportunity to concentrate on these areas for an extended period. In particular, his goal was to study the use of image color variation for creating keypoint descriptors that could aid in recognizing objects in images with varying camera viewpoint and lighting.
Professor Olson might have been on sabbatical, but he still included multiple students in his research. In collaboration with Siqi Zhang (UW Bothell CSSE undergraduate), Dr. Olson was able to show that color can be used in conjunction with shape in order to improve object recognition performance in the context of image keypoint descriptors. They further demonstrated, with the help of Sam Hoover and Jordan Soltman (also, UW Bothell CSSE undergraduates), that the combination of several complementary keypoint descriptors (including our color descriptor) could improve upon any single keypoint descriptor, including those based on recent work in deep learning. Now that Dr. Olson’s sabbatical is over, he plans to explore the application of these ideas to image classification and to further study deep learning techniques for these problems.
Professor Olson’s sabbatical experience helped grow a new research thread; he anticipates exploring this research thread for several years to come. This sabbatical experience has already positively influenced his teaching in CSS 487 (Computer Vision), where he included discussion of the newly developed keypoint matching techniques. CSS undergraduate and graduate student can contribute to the continued exploration of Dr. Olson’s research and future work, including capstones, research projects, and Master’s theses.
MARC DUPUIS: TVW’S Interview on Election Security
In October, Assistant Professor Marc Dupuis from the Computing & Software Systems Division, was interviewed on TVW’s The Impact. The interview ranged from security concerns regarding November’s election to the growing threat of hacking personal information.
“Cyber security is at a heightened level than what we have seen in the past,” Professor Dupuis stated. As new technologies emerge, threats will adapt—forcing us to always play catch-up. “Hackers only need to be right once, we have to be right one-hundred percent of the time,” says Dr. Dupuis. It’s becoming increasingly important to secure systems as hackers become more sophisticated.
Dr. Dupuis talked about how there are four different types of hacker groups, each with their own level of sophistication and capabilities. “Instead of worrying about robbing a bank or doing something where they can easily get caught, cyber-criminal activity on the internet is much easier in many respects—the risk is a lot lower, and the payouts a lot of the time are greater,” according to Professor Dupuis.
“Part of the challenge” Dr. Dupuis stated, “is that security is never someone’s first priority. They want to do something, engage by checking their email or though social networking. But for the average everyday person, if you do what you can do you’ll generally be well protected.” Dr. Dupuis recommends you take three key steps to help protect yourself from persistent threats:
Have good backups of all your important files, including one hardware version ‘off-site’
Make sure your antimalware software is current and up-to-date
Use a password manager