HISTON-BASED SEGMENTATION FOR SEMANTIC SCENE CLASSIFICATION AND OBJECT LABELING USING DYNAMIC FEATURES MODELING

Abstract

Visual interpretation of the scenes precisely requires visionary information on the detection of objects and scene types. The visual interpretation and understanding are fundamental tasks for many applications, such as robotic vision, image-based modeling and augmented reality scene integration. In this paper, a histon based segmentation model is designed that can partition each object into separate regions. Then, the dynamic features of those regions are extracted that extends the key points features, ridge detection and difference of Gaussian histograms. The combination of these features provide a reasonable fact to detect the objects in the scene. Labeling of these objects are performed on the basis of similar features. Finally, the recognizer engine is used to recognize different complex scenes. The experimental results show a better performance on 15 Scene and PASCAL VOC datasets.