Software created by Georgia Tech researchers proved crucial during a recent project to devise an imaging system that could someday allow historians to read fragile antique books without turning the pages.
At its most basic level, the software — created by Alireza Aghasi, a former postdoctoral researcher at Georgia Tech who now works at the Massachusetts Institute of Technology (MIT), and Justin Romberg, Schlumberger Professor in Georgia Tech’s School of Electrical and Computer Engineering — recognizes shapes and patterns in images.
But the software is so powerful it can pick out shapes even a human eye cannot, even in images that are extremely noisy or distorted.
Aghasi and Romberg partnered with researchers at MIT on a terahertz imaging system that uses timed bands of electromagnetic radiation to read letters printed on a stack of pages. The system is able to distinguish letters on different pages based on the time it takes for each image to be reflected back.
The noisy images are then processed by the shape recognition software, which strings letters into words and words into sentences.
The shape processing software starts by comparing patterns in a noisy photograph to a database of shapes, looking for matches. The repository uses simple shapes such as circles and squares to create much more complicated shapes and combinations.
For example, to arrive at an omega symbol, the database combines two circles of different sizes, a rectangle, and a triangle. The algorithm knows what pieces of each shape make up the omega and uses those to match and record what it sees in the grainy images.
The project with MIT is just the first step of customizing the software to work with other technologies. Another project in the works is looking at how to decipher pages that are bent or distorted. Outside of reading closed books, the researchers say, the software could be adapted for a range of uses. — Josh Brown