Hybrid Region Merging For Image Segmentation Using Optimal Global Feature With Global Merging Criterion Approach

• Main Author: Vadiveloo, Mogana
• Format: Thesis
• Language: English
• Published: 2020
• Subjects: QA75.5-76.95 Electronic computers. Computer science
• Online Access: http://eprints.usm.my/55616/1/Pages%20from%20THESIS%20by%20MOGANA%20cut.pdf; http://eprints.usm.my/55616/
• Institution: Universiti Sains Malaysia

Region merging approach is used to reduce over segmented regions produced by region-based image segmentation algorithms. It is performed by merging the over segmented regions progressively to produce the final segmentation as spatially contiguous regions with closed boundaries. Predominantly, region merging is performed between two neighboring regions solely on a local merging criterion. This may fail most existing region merging approaches to detect large non-homogeneous visual objects that have global semantic similarity but consist of diverse set of over segmented regions. Besides that, improper selection of global feature information by partitional clustering algorithm in turn affects the merging criterion derivation in region merging eventually causing leakages into adjacent visual object regions. Consequently, this thesis aims to solve these two issues by proposing a region merging approach to merge the over segmented regions producing semantic segments of visual objects regions.