Hierarchical Objects Semantic Graph Based Hybrid Learning Method for Automatic Complicated Objects Recognition
Sun Xian Fu Kun Wang Hong-qi
(Institute of Electronic, Chinese Academy of Sciences, Beijing 100190, China)
(Key Laboratory of Spatial Information Processing and Application System Technology,Chinese Academy of Sciences, Beijing 100190, China)
Abstract：Automatic objects recognition is a key issue in image processing area. A new hierarchical objects semantic graph based hybrid learning method is proposed to recognize targets in complicated images. This method builds a hierarchical semantic graph model to reinforce the semantic constraints among targets, background, and components in images. And it also proposes a belief objects network to improve the utilization of spatial information, by using local classifier to calculate objects properties and using belief messages to propagate the objects relationships. Besides, the method uses discriminative learning and generative learning interleavely to improve the training error, memory usage and recognition efficiency. Experimental results demonstrate that the proposed method is meaningful and helpful for image understanding.
2011, Vol. 33 
