A Threshold Selection Method from Gray-Level Histograms¶
Why this mattered¶
Otsu’s 1979 paper mattered because it turned image thresholding from a largely heuristic preprocessing choice into a simple, reproducible statistical optimization problem. By selecting the gray-level threshold that maximizes between-class variance, the method gave practitioners an automatic way to separate foreground from background using only the image histogram. This was especially important for document analysis, microscopy, industrial inspection, and early machine vision, where binarization was often the first step before measurement, recognition, or segmentation.
The paradigm shift was not that thresholding itself was new, but that Otsu showed a broadly usable criterion could be computed efficiently and without hand-tuned parameters. After this paper, threshold selection became a standard baseline: easy to implement, fast enough for routine use, and mathematically transparent. Its success helped establish histogram-based image segmentation as a disciplined part of digital image processing rather than an ad hoc craft.
Otsu’s method also became a point of comparison for later breakthroughs in computer vision. More advanced segmentation approaches, including adaptive thresholding, region-based methods, clustering, graph cuts, and eventually learning-based segmentation, often addressed cases where Otsu’s global bimodal assumption was insufficient. Precisely because it was so simple and robust, the method remained influential: it defined a canonical automatic-thresholding problem and supplied a benchmark that later methods had to surpass.
Abstract¶
(no abstract available)