Physics-Based Feature Engineering

Abstract

We describe a new paradigm in computational imaging for performing edge and texture recognition with a superior dynamic range compared to other methods. The algorithm has its origin in Photonic Time Stretch, a realtime instrumentation technology that has enabled observation of ultrafast, non-repetitive dynamics and discovery of new scientific phenomena. In this chapter, we introduce the mathematical foundation of this new transform and review its intrinsic properties including the built-in equalization property that leads to high performance in visually impaired images. The algorithm is spearheading the development of new methods for feature engineering from visually impaired images with unique and superior properties compared to conventional techniques. It also provides a new approach to the computation of mathematical derivatives via optical dispersion and diffraction. The algorithm is a reconfigurable mathematical operator for hyper-dimensional feature detection and signal classification. It has also shown promising results in super-resolution single molecule imaging.