Image Categorization of Integrated Generative Models and Discriminative Methods
GUO Li-jun;ZHAO Jie-yu;SHI Zhong-zhi
1.Institute of Computer Technology,CAS,Beijing,100080,China;2.Faculty of Information Science and Engineering,Ningbo University,Ningbo,Zhejiang 315211,China;3.Graduate University of Chinese Academy of Sciences,Beijing 100039,China
In our approach,the global visual vocabulary which is similar to keypoints of codebooks is built with Gaussian Mixture Models based on local image features.Images are represe nted as a new set of feature vectors which are summed posteriori responsibility relative to different visual words.The discriminative classifier is trained by Support Vector Machine with linear kernels based on above features.Experiments were performed on the PASCAL VOC 2006 dataset and the results suggested the influence of background factors on classification effectiveness.And further experiments showed that the features extracted from object areas can be combined effectively to improve classification performance in our method.
2010, Vol. 38 
