People often talk about making a difference in the world, but Thomas Tsao and James Zhiqing Wen are concerned with seeing the differences in the world. The two worked together on an image tracking system modeled on the biological vision system, which perceives changes in a series of images instead of searching for common elements. Thus, it is a tracking system based on invariants instead of constants.
While both were working at CompuSensor Technology Corp. (CST; Gaithersburg, MD), Tsao and Wen wrote a paper, "Image-based target tracking through rapid sensor orientation change," that focused on this concept. The paper was published in the March 2002 issue of SPIE's journal Optical Engineering and received the 2002 Rudolf Kingslake Medal and Prize, which is awarded annually to recognize the most noteworthy original paper to appear in Optical Engineering on the theoretical or experimental aspects of optical engineering.
James Zhiqing Wen seeing the same thing
The biological vision system has fascinated scientists and engineers for decades. Following or tracking an object with the eyes as it moves across a varied background with changing light and clarity seems to be so easy for biological systems. However, reproducing this concept artificially has been a colossal challenge.
Thomas Tsao, right, and his daughter, Doris, left, working together in the lab.
"The old way of thinking is to find something in common between the different images to support the decision of 'seeing the same thing,'" explains Tsao. "Researchers have tried various methods to find the same image contents based on the idea of computing the correlation and/or identifying for special common features such as edge polarity, size, color, brightness, etc., or some abstract invariants represented by constants extracted from the same scene area. It is natural to think that in order to track something, you have to first identify the same thing across the entire image sequence. But the biological vision system does not work that way."
Human vision, for instance, works on the basis that everything is constant or stable and is only concerned with changes in the scene instead of similarities. "We see 'the same thing' not because we find some constants or something in common, but because we can successfully factor off the transformations. Seeing 'the changes' actually facilitates the process of seeing the 'same thing,'" says Tsao, who has developed his theory and computational method over the past 18 years. testing the theory
Realizing this aspect of biological vision systems, Tsao began work on a mathematical method based on his theory, which focused on computing geometric transformations. Using Gabor functions as a medium for local geometric transformations, Tsao implemented Lie transformation group machinery, allowing a smooth, responsive mathematical method.
Using this method, Wen then played a key role in the software development, implementation, test, and evaluation of the image-based biological target-tracking algorithm. "I tuned and refined the algorithm to ensure the speed, accuracy, and robustness for the realistic target-tracking tasks raised by the U.S. Army," says Wen.
Wen's broad range of research interests include optical and digital signal processing, artificial intelligence, fuzzy logic, machine vision, and optical metrology, to name just a few. "During my graduate study I worked on projects to solve problems facing optoelectronic implementation of neural networks," says Wen. "After I joined CST, I learned more about the neurobiological discoveries in vision systems. The capability and simplicity behind the biological vision system gave me great motivation to work with Dr. Tsao to develop and implement the artificial vision system that closely mimics the biological vision system for pattern recognition and target tracking."
This artificial vision system has many applications beyond the missile tracking example used in their paper in Optical Engineering. In fact, this technology can be used in applications as varied as robotic vision for quality control, 3-D modeling in the movie industry, real-time visual-based tracking in a cluttered road traffic scene, and unmanned aerial vehicle sensor data processing.
Since writing the paper, Tsao has continued to further develop the theories and methods covered in the paper at CST. Wen, however, has since joined Accumux Technologies, Inc. (Camarillo, CA), where he led the development of a birefringent interleaver project. He also played an important role in the development of the tunable chromatics dispersion compensator, which won the Lightwave "Best in Show" at the National Fiber Optic Engineers Conference in 2002.
"Humans so far only have a limited understanding of the secrets and mechanisms inside biological intelligent systems, including imaging or vision systems," says Wen. "I still have a great interest in this area. I hope to create some practical optoelectronic and computer systems and devices that are as smart as or even, to some extent, better than the biological systems."
Wen and his wife Judy Xiaoyun Zhang live in California with their two daughters, Nancy, a fourth grader, and Lucy, who is going into kindergarten this fall.
All in the family
The saying, "the apple doesn't fall far from the tree," is an apt description of the Tsao family. Tsao's wife Susan Chang has an MS in computer vision from the University of Maryland (College Park, MD) and works for the Naval Surface Warfare Center, Carderock Division (West Bethesda, MD) on the acoustic signal processing tutoring system. And their two children, Albert and Doris, certainly haven't fallen far from the tree.
Albert will be a junior in high school this fall in a science magnet program, and Doris has a BS in biology and mathematics from the California Institute of Technology (Pasadena, CA) and a PhD from Harvard Medical School (Boston, MA).
"For a long time, my daughter and I wanted to systematically develop a geometric theory of visual perception," says Tsao. "Last year, she finished her PhD under the supervision of Margaret Livingston, and is now a post-doc under Professor David Hubel, a Nobel laureate, and both are authors I quoted in my paper. Using fMRI, she has identified the area in the macaque brain that is most crucial for processing surface information and is now studying the physiology of this area in detail. Doris has also spent a substantial amount of time studying modern geometry, and we are collaborating to figure out the circuits that the brain uses to represent the invariant geometry of the environment."
Tsao mentions among his hobbies that he "enjoys discussing science and mathematics problems with my kids." Those are sure to be some lively discussions.
-Erin M. Schadt