Abstract
In a microscopic image, fiber cross sections are often surrounded by borders distinctively darker than their bodies and the background. Fiber borders can be utilized to separate cross-sections properly so that accurate fiber shape and size information can be obtained. Hence, locating correct fiber borders is one of the most critical steps in cross-sectional analysis for fiber characterization and identification. This paper introduces a dual-thresholding algorithm that performs automatic fiber border segmentation from noisy cross-sectional images. The dual thresholds include a low threshold calculated based on the histogram of the difference from the average grayscale, and a high threshold computed by a bisection algorithm. With the low threshold, part of fiber border pixels, regarded as seeds, can be reliably located. The seeds can be further expanded by using the high threshold to form complete borders surrounding individual cross-sections. The experimental results show that the dual-thresholding algorithm can obtain cleaner and more fiber borders than other connectional thresholding algorithms, and improves the detection accuracy from 52.78% and 88.88%.