Dynamic sampling for scanned microscopy modalities

Abstract

We present an unsupervised heuristic-based batched dynamic algorithm for pixel-wise image sampling in a variety of scanned imaging modalities. Our algorithm offers improved performance over static Sobol sampling at a low computational cost. We evaluate the algorithm using three datasets to highlight potential savings in both two and three dimensional characterization contexts. Significantly, the implementation speed lends the approach runtime compatibility with electron microscopy. Reconstructed grain shape distributions for the sampled data sets are shown to be close to those of the original microstructures.