Super-resolution microscopy using normal flow decoding and geometric constraints.

Abstract

Prior knowledge about the observed scene provides the key to restoration of frequencies beyond the bandpass of an imaging system (super-resolution). In conjunction with microscopy two super-resolution mechanisms have been mainly reported: analytic continuation of the frequency spectrum, and constrained image deconvolution. This paper describes an alternative approach to super-resolution. Prior knowledge is imposed through geometric and dynamic models of the scene. We illustrate our concept based on the stereo reconstruction of a micropipette moving in close proximity to a stationary target object. Information about the shape and the movement of the pipette is incorporated into the reconstruction algorithm. The algorithm was tested in a microrobot environment, where the pipette tip was tracked at sub-Rayleigh distances to the target. Based on the tracking results, a machine vision module controlled the manipulation of microscopic objects, e.g. latex beads or diamond mono-crystals. In the theoretical part of this paper we prove that knowledge of the form 'the pipette has moved between two consecutive frames of the movie' must result in a twofold increase in resolution. We used the normal flow of an image sequence to decode positional measures from motion evidence. In practice, super-resolution factors between 3 and 5 were obtained. The additional gain originates from the geometric constraints that were imposed upon the stereo reconstruction of the pipette axis.