Accurate measurement and identification of fibers using magnified digital images relies mainly on the quality of the fiber image. At a given power of magnification, a light microscope has a limited depth of field that may not cover the entire depth space of a fiber sample on the slide and thus disallows all fibers in the image from being well focused, regardless of focus positions. This paper introduces an image-fusion technique to solve mal-focused fibers in a microscopic image to ensure optimal image quality for fiber measurements. This new technique utilizes multiple images of the same view taken at consecutive depths, calculates a focus measure of every pixel in each image, and constructs a matrix to register the image layer that has the maximum focus measure for every pixel. The matrix can be further modified and then used as a map to reconstruct a new image that contains only the best-focused pixels out of the captured images. The fused image combines selected features of multi-focus images so that unfocused fibers can be realistically amended and blurring fiber edges can be sharpened. Compared to the data measured from a single-focus image, the data taken from the fused image can greatly improve the accuracy of fiber thickness measurements.

The Euclidean distance transform is one of the fundamental operations in image processing. It has been widely used in computer vision, pattern recognition, morphological filtering, and robotics. This paper proposes a systolic algorithm that computes the Euclidean distance map of an N x N binary image in 3N clocks on 2N(2) processing cells. The algorithm is designed so that the hardware resources are reduced; especially no mulitipliers are used and, thus, it facilitates VLSI implementation.

Fiber cross-sectional shapes can influence many physical properties of fibers. Automated identification of shaped fibers is critically important for fiber quality inspection. This paper presents a distance-based skeletonization algorithm used for reliable identification of shaped fibers. The skeleton of a fiber cross section, which is generated from fiber distance maps and maximal disks, is ensured to be continuous and insensitive to edge noise, and therefore can be used as abstract representations of fiber topology for shape analysis. A set of shape descriptors are defined from fiber skeletons and a support vector machine method is used to classify fibers based on the shape measurements. The experimental results show that the presented approach can be used to recognize shaped fibers based on the analysis of skeleton structures.

No abstract

No abstract