Hierarchical image segmentation using local and adaptive similarity rules

Segmentation techniques applied to remotely-sensed data can increase the accuracy of object extraction, particularly for scenes of natural homogeneous regions such as agricultural sites. However, the accuracy becomes very low when the natural object's boundaries are not well defined. This paper presents a modified approach to hierarchical image segmentation by step-wise optimization developed by J. M. Beaulieu. It uses local similarity rules to extract homogeneous objects from remotely-sensed data. A modified Student-t test based adaptive criterion is introduced to define automatically the final number of segments. These rules and tests represent the main contribution to the basic algorithm.