Publication: Image segmentation algorithm based on integration of k-means clustering, watershed and binary partition tree
Date
2012-06-01
Authors
Wong, Kok Choy
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Nowadays, computer vision technology had grown at a supreme speed. Along with the advancement of computer vision technology, many computer scientists had focused in the development of image processing algorithm. With the aim to construct automated feature detection and information extraction image processing algorithm, the image segmentation is being explored by many researchers. Although many types of image segmentation technique had been proposed, their accuracy and efficiency are still far away from detection done by human, especially when the image experiences some illumination variations (Uneven lighting, light reflection, etc.). Thus, the main objective of this project is to improve the robustness of an image segmentation algorithm against various illumination conditions. The improved image segmentation algorithm presented here is the use of K-Means Clustering algorithm as presegmentation, and its output will undergo Watershed Transform before Binary Partitioning Tree (BPT) merging process takes place. K-Means Clustering is implemented to reduce illumination changes, and its output image undergoes Watershed Transform. The resultant regions obtained from Watershed Transform are then used as the leaves node in BPT. Based on merging criteria, these regions will be merged two by two until it reaches the root of the tree (the entire image). To evaluate the performance of the proposed algorithm, images and ground truth result in Sharon Alpert’s segmentation database are used to evaluate the illumination compensation effect. From the result obtained we can say that by adding K-Means Clustering algorithm, the segmentation process is now more robust against illumination variation.