Output details

Image segmentation based on semi-greedy region merging

<23> Originality: This paper proposes a practical solution that increases the speed of segmentation while allowing for global feature cohesion. This is achieved by performing multiple region merging operations in each iteration and dynamic merge relaxation for feature discrimination.

Significance: This algorithm forms the first step of a framework for semantic visual understanding that considers unsupervised feature extraction, a matching feature scheme and conceptual categorization of classes.

Rigour: Extensive analysis of the results through benchmark databases. The work forms part of a PhD thesis completed in 2013, and has formed the basis for presentations in IJCN 2012 and ICIP 2012.