To develop complex algorithms enabling optical system calibration, image segmentation, object detection, and classification to optimize the Picodya imaging system’s function and its B-Matrix image content.
Combining our computer vision capabilities, machine Learning proficiency, and system-level experience with Picodya’s engineering expertise, we co-developed an on-site calibration procedure that enables inter-device repeatability and periodical testing. Our machine learning classification module was designed to distinguish between different bio-markers concentration levels and account for image source instability, high-dynamic-range scenes, optical aberrations, and photon count estimation. We designed specialized optimization cost functions for detection, registration, and feature extraction modules in cases where biological markers outliers appear in the image. In addition, numerical optimization procedures built into the system account for location uncertainty, orientation instability, and intensity variability.