Image Segmentation Using Normalized Cuts and Efficient Graph-Based Segmentation

In this paper we propose an hybrid segmentation algorithm which incorporates the advantages of the efficient graph based segmentation and normalized cuts partitioning algorithm. The proposed method requires low computational complexity and is therefore suitable for real-time image segmentation processing. Moreover, it provides effective and robust segmentation. For that, our method consists first, at segmenting the input image by the "Efficient Graph-Based" segmentation. The segmented regions are then represented by a graph structure. As a final step, the normalized cuts partitioning algorithm is applied to the resulting graph in order to remove non-significant regions. In the proposed method, the main computational cost is the efficient graph based segmentation cost since the computational cost of partitioning regions using the Ncut method is negligibly small. The efficiency of the proposed method is demonstrated through a large number of experiments using different natural scene images.