Superpixel level object recognition under local learning framework

In this paper, we propose a simple yet effective method for superpixel level object recognition on the bag-of-feature framework. Instead of using general classifiers for the superpixel categorization, we introduce local learning classifiers into our framework, which aims to turn a highly non-linear classification problem into multiple local linear problems within different subsets of the database, so as to tackle the intraclass variation problem brought by superpixel based representations of objects. In addition, context information is used to make better performance by combining each superpixel with its appearance-based superpixel neighbors within a certain neighborhood distance from superpixel mean color map. At last, we utilize superpixel based Graph Cuts algorithm to segment the objects from background image. We test the proposed method on Graz-02 dataset, and get results comparable to the state-of-the-art.