Date on Master's Thesis/Doctoral Dissertation
Applied and Industrial Mathematics, PhD
Committee Co-Chair (if applicable)
Sahoo, Prasanna (Deceased)
image segmentation; image thresholding; two-dimensional histogram; entropy; particle swarm optimization
In this dissertation, we discuss multi-level image thresholding techniques based on information theoretic entropies. In order to apply the correlation information of neighboring pixels of an image to obtain better segmentation results, we propose several multi-level thresholding models by using Gray-Level & Local-Average histogram (GLLA) and Gray-Level & Local-Variance histogram (GLLV). Firstly, a RGB color image thresholding model based on GLLA histogram and Tsallis-Havrda-Charv'at entropy is discussed. We validate the multi-level thresholding criterion function by using mathematical induction. For each component image, we assign the mean value from each thresholded class to obtain three segmented component images independently. Then we obtain the segmented color image by combining the three segmented component images. Secondly, we use the GLLV histogram to propose three novel entropic multi-level thresholding models based on Shannon entropy, R'enyi entropy and Tsallis-Havrda-Charv'at entropy respectively. Then we apply these models on the three components of a RGB color image to complete the RGB color image segmentation. An entropic thresholding model is mostly about searching for the optimal threshold values by maximizing or minimizing a criterion function. We apply particle swarm optimization (PSO) algorithm to search the optimal threshold values for all the models. We conduct the experiments extensively on The Berkeley Segmentation Dataset and Benchmark (BSDS300) and calculate the average four performance indices (Probability Rand Index, PRI, Global Consistency Error, GCE, Variation of Information, VOI and Boundary Displacement Error, BDE) to show the effectiveness and reasonability of the proposed models.
Surina, Surina, "Information theoretic thresholding techniques based on particle swarm optimization." (2018). Electronic Theses and Dissertations. Paper 3040.
Retrieved from https://ir.library.louisville.edu/etd/3040