Region Labeling in Natural Scene Images

Semantic region labeling is important in the fields of low, mid and high level computer vision. This paper proposes an approach for region labeling in natural scene images using over-segmented regions as a basic unit. This system consists of three phases: segmentation, feature extraction and labeling. The paper uses Marker-controlled Watershed Segmentation (MCWS) algorithm for segmented regions generation. The aim of this system focuses labeling on nine semantic concept classes using color, texture features and 3-layer Feed Forward Neural Network (FFNN) classifier. The system performance is evaluated on the use of public MSRC 9class dataset.