Unattenuated tracer particle extraction through time-averaged, background image subtraction with outlier rejection

Abstract Fluid-flow analysis using particle tracking seeks to assign velocity vectors to sequences of still images (tracks) of particles suspended in a transparent fluid or gas. This requires that high quality particle images be obtained from a system of moving particles. In practice, however, the images are contaminated by a variety of noise sources which must be removed before tracking can be performed. The traditional approach to prefiltering, which is being used in commercially-available systems, is to perform background subtraction in concert with some form of thresholding and/or image stretching. Unfortunately, these methods can attenuate particle images so badly that valid track yields are significantly reduced. In place of these methods, we present a non-attenuating background subtraction method with outlier rejection together with a non-attenuating substitute for thresholding.These algorithms have been tested on real track data and can recover virtually all images of particles in suspension with very little attenuation of particle-image intensity.