Comparison of Perceptual Grouping Criteria within an Integrated Hierarchical Framework

摘要

The efficiency of a pyramid segmentation approach mainly depends on the graph selected to encode the information within each pyramid level, on the reduction or decimation scheme used to build one graph from the graph below, and on the criteria employed to define if two adjacent regions are similar or not. This paper evaluates three pairwise comparison functions for perceptual grouping into a generic framework for image perceptual segmentation. This framework integrates the low-level definition of segmentation with a domain-independent perceptual grouping. The performance of the framework using the different comparison functions has been quantitatively evaluated with respect to ground-truth segmentation data using the Berkeley Segmentation Dataset and Benchmark providing satisfactory scores.