Mark Gesley and Romin Puri, Spynsite LLC, Oakland, CA
ABSTRACT
Objects of interest are rendered from spectral images. Seven types of blood and cancer cells are imaged in a micro-scope with changes in source illumination and sensor gain over one year calibrated. Chromatic distortion is measured and corrections analyzed. Background is discriminated with binary decisions generated from a training
sample pair. A filter is derived from two sample-dependent binary decision parameters: a linear discriminant and a minimum error bias. Excluded middle decisions eliminate order-dependent errors. A global bias maximizes the number and size of spectral objects. Sample size and dimensional limits on accuracy are described using a
covariance stability relation.
Journal of the Optical Society of America A Vol. 39, No. 11, November 2022, Pp. 2035-2044.