High speed method for in situ multispectral image registration

Abstract

High speed data registration is required for the study of fluorescence resonance energy transfer in real time as well as fast dynamic intra- and inter-cellular signaling events. Multispectral confocal spinning disk microscopy provides a high resolution method for performing such real time live cell imaging. However, optical distortions and the physical misalignments introduced by the use of multiple acquisition cameras can obscure spatial information contained in the captured images. In this manuscript, we describe a multispectral method for real time image registration whereby the image from one camera is warped onto the image from a second camera via a polynomial correction. This method provides a real time pixel-for-pixel match between images obtained over physically distinct optical paths. Using an in situ calibration method, the polynomial is characterized by a set of coefficients, using a least squares solver. Error analysis demonstrates optimal performance results from the use of cubic polynomials. High-speed evaluation of the warp is then performed through forward differencing with fixed-point data types. Forward differencing is an iterative approach for evaluating polynomials on the condition that the function variable changes with constant steps. Image reconstruction errors are reduced through bilinear interpolation. The registration techniques described here allow for successful registration of multispectral images in real time (exceeding 15 frame/s) and have a broad applicability to imaging methods requiring pixel matching over multiple data channels.