TY - JOUR
AU - Saman Sarraf
PY - 2017/02/15
Y2 - 2020/09/29
TI - Binary Image Segmentation Using Classification Methods: Support Vector Machines, Artificial Neural Networks and Kth Nearest Neighbours
JF - International Journal of Computer (IJC)
JA - IJC
VL - 24
IS - 1
SE - Articles
DO -
UR - https://www.ijcjournal.org/index.php/InternationalJournalOfComputer/article/view/832
AB - The principal objective of this work is to demonstrate efficient parameter selection for various networks used in binary image segmentation. The Support Vector Machines using four kernel functions (i.e., Radial Basis Function, Quadratic, Polynomial, and Linear), Neural Networks (i.e., Feed-forward Back-propagation) and Kth Nearest Neighbours algorithm were applied to five different datasets that had been generated from a given image. Pixel coordinates (x,y) were considered as inputs. Grid search and cross-validation were performed to identify the optimal network parameters. All experiments were repeated five times in order to develop confidence in the obtained results. High accuracy was achieved in most cases 95% for SVM-RBF, 90.4% for SVM-Quadratic, 90.8% for SVM-Polynomial, 60% for SVM-Linear, 88% for Neural Networks and 97% for K-NN. After grid search for SVM-RBF, the accuracy reached 98%. In this project, SVM-RBF showed a high level of accuracy and consistency. It was also found that the selected features (pixel coordinates) were discriminative.
ER -