A0572
Title: Bit-plane probability model and its application in image segmentation
Authors: Kwun Lun Chu - The Hang Seng University of Hong Kong (Hong Kong) [presenting]
Abstract: The task of image segmentation is complex in computer vision and has a broad range of applications in fields such as medical imaging and scene analysis. One popular approach often starts with dividing images into non-overlapping blocks and grouping them by similar image features, resulting in an initial segmentation. Boundary correction is then applied to improve this segmentation. Under this framework, selecting the grouping methodology and image features is crucial to the performance of the segmentation result. Traditional approaches in the literature often adopt certain clustering algorithms and rely on the histograms of the image. However, these approaches are very sensitive to initial clusters, and the number of clusters is often unknown, while histograms can be inefficient due to their high dimensionality. To address these challenges and enhance image segmentation accuracy, a novel probability model is proposed to characterize the distributions of image variations based on bit-plane probabilities and dependencies, providing a universal parametric representation that can model any random distribution. In addition, the mathematical optimization framework integrates this model with an agglomerative fuzzy k-means algorithm, incorporating spatial information and a pixel-based boundary localization algorithm for different image segmentation applications. The experimental results demonstrate that the method outperforms current state-of-the-art approaches.
