Natural Scene Segmentation Method through Hierarchical Nature Categorization

摘要

In this paper we present a hierarchical learning method to segment natural colour images combining the perceptual information of three natures: colour, texture, and homogeneity. Human knowledge is incorporated to a hierarchical categorisation process, where each nature features are independently categorised. Final segmentation is achieved through a refinement process using the categorisation information from each segment. Experiments are performed using the Berkeley Segmentation Dataset achieving good results even when comparing them to other significant methods.