Scene categorization based on object bank

Abstract

Scene Categorization is one of the most competitive topic in Robotics and computer vision. It classifies the given image based on the scene information. Object Bank is an object-level image representation for high-level visual recognition. Compared with low-level feature representations, Object Bank offers more rich description of images. In this paper, we proposed a Scene Categorization method based on Object Bank. This method improved the Object Bank by reducing the dimensions of the feature vectors extracted from the image and add low-level representation Locality-constrained Linear Coding. The experiments show that the classification efficiency and classification accuracy of the improved method is much higher.