Fast autofocus algorithm for automated microscopes.

Abstract

We present a new algorithm to determine, quickly and accurately, the best-in-focus image of biological particles. The algorithm is based on a one-dimensional Fourier transform and on the Pearson correlation for automated microscopes along the Z axis. We captured a set of several images at different Z distances from a biological sample. The algorithm uses the Fourier transform to obtain and extract the image frequency content of a vector pattern previously specified to be sought in each captured image; comparing these frequency vectors with the frequency vector of a reference image (usually the first image that we capture or the most out-of-focus image), we find the best-in-focus image via the Pearson correlation. Numerical experimental results show the algorithm has a fast response for finding the best-in-focus image among the captured images, compared with related autofocus techniques presented in the past. The algorithm can be implemented in real-time systems with fast response, accuracy, and robustness; it can be used to get focused images in bright and dark fields; and it offers the prospect of being extended to include fusion techniques to construct multifocus final images.