Region-based image segmentation

Abstract

The purpose of this thesis is developing the algorithms of region based image segmentation. The region growing operates in one pass and groups the pixels spatially adjacent (proximity) and similar in characteristics (similarity). During grouping operation, the threshold values are allowed to vary with respect to the variance of the each region. So the region growing can actively cope with the complexity of the image array. Since the region grower mainly focus attention on the local features of the image array, they have many weakness in the global point of view. So I use simple heuristics to segment the objects and background. The merged region lists are used to construct hierarchical tree structure for emphasizing the focus of attention returned from high level part.