Image recognition has become increasingly critical in applications ranging from smartphones to driverless cars, and on Wednesday UCLA opened up to the public a new algorithm that promises big gains.
The Phase Stretch Transform algorithm is a physics-inspired computational approach to processing images and information that can help computers “see” features of objects that aren’t visible using standard imaging techniques. It could be used to detect an LED lamp’s internal structure, for example—something that would be obscured to conventional techniques by the brightness of its light. It can also distinguish distant stars that would normally be invisible in astronomical images, UCLA said.
Essentially, the algorithm works by performing a mathematical operation that identifies objects’ edges and then detects and extracts their features. It can also enhance images and recognize objects’ textures. Potential applications include face, fingerprint and iris recognition for high-tech security as well as navigation systems in self-driving cars or industrial-product inspection.
Developed by a group led by Bahram Jalali, a UCLA professor of electrical engineering, and senior researcher Mohammad Asghari, the Phase Stretch Transform algorithm grew out of UCLA research on a technique called photonic time stretch, which has been used for ultrafast imaging and detecting cancer cells in blood. By open-sourcing its code, the researchers hope to enable others to incorporate the technology into computer-vision, pattern-recognition and other image-processing applications, they said.
It’s now available on GitHub and the MATLAB Central File Exchange.