Date of Award


Degree Type


Degree Name

Master of Science


Mechanical Engineering

Major Professor

Ahmad Vakili

Committee Members

Basil Antar, Louis Deken


The University of Tennessee Space Institute is interested in pitch-based carbon fibers. A need has arisen for characterizing relative alignment of nonwoven fiber mats, produced by the spinning gathering technology implemented. Knowledge of this alignment is relevant in the validation of production methodology as well as aiding in the analysis of carbon fiber composite materials made with the fiber mats. A digital image processing method was developed presented and evaluated in this thesis consisting of three steps: preprocessing using image smoothing, edge detection based upon local intensity gradients, and line detection via the Hough Transform. A summary of typical results are provided in both Chapter 4; a more extensive examples of analysis as well as Matlab™ codes are provided in the Appendix. This method was applied to digital photographs taken from two sources: a carbon fiber mat produced by the University and a generic anti-static dryer sheet.

Four preprocessing methods are explored: Averaging, Median, Gaussian, and Laplacian of Gaussian smoothing. In each case, a convolution kernel is created based upon specified dimensions and standard deviation σ (for Gaussian and Laplacian of Gaussian). Original images are convolved with each kernel, producing a smoothed image. Noise reduction is quantified by the difference between the original image and the smoothed image; this noise is analyzed in histograms generated from Matlab™. Using parameters such as mean, median, and maximum grayscale value as well as processing time, a 5 x 5 Averaging kernel is identified as the optimum preprocessing method explored.

In this thesis, three edge detection techniques are evaluated: the Roberts Method, the Sobel Method, and the Canny Method. Roberts and Sobel, considered first-order methods, are evaluated solely over a specified range of thresholds. The second-order Canny Method (with standard deviation σ = 2.75) varied over a range of thresholds of a higher order of magnitude. Histograms generated from binary edge images, as well as processing time and relative error introduced from image orientation are used for evaluation purposes. The Canny Method, using a threshold interval of (0.010, 0.250) is chosen as the most appropriate method for this analysis.

The Hough Transform is used to determine characteristics of the lines found in the edge images. Each line is translated into Hough-space using parametric terms ρ and Θ. Using built-in Matlab™ functions, each edge image is divided into grids ranging from 1 x 1 to 10 x 10. In each grid, the mean angle θ, total length L, and processing time t are determined. Results are returned in an array of the specified dimensions.

Files over 3MB may be slow to open. For best results, right-click and select "save as..."