Abstract：The discriminative models are sensitive to limited training samples, which usually has poor generalization performances and is easily over-fitting. A hybrid discriminative approach with Bayesian prior constraints is proposed to solve this issue.By introducing the generative prior analysis into the discriminative approach, a complementary learning structure is built to fuse different classification results.The different types of classifiers are trained separately, and an effective fusion decision is defined to obtain the most confident testing samples along with the estimated labels. By enlarging the training set automatically, the model is updated to make up for the incomplete distribution information of training samples.The experimental results show that compared with the classical methods, the proposed approach can effectively update the model by figuring out the discriminating samples and correct the misclassifications caused by the uneven distribution of limited samples. It can improve the performances of scene categorization.
[1] Benenson R, Mathias M, Timofte R, et al. Pedestrian Detection at 100 Frames per Second // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Providence, USA, 2012: 2903-2910 [2] Zhang L, Van Der Maaten L. Structure Preserving Object Tracking // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Portland, USA, 2013: 1838-1845 [3] Wang Q Z, Kang W W, Wang B. Design of 3D Latent-SVM and Application to Detection of Lesions in Chest CT. Pattern Recognition and Artificial Intelligence, 2013, 26(5): 460-466 (in Chinese) (王青竹,康文炜,王 斌.三维隐SVM算法设计及在胸CT图像病灶检测中的应用.模式识别与人工智能, 2013, 26(5): 460-466) [4] Wang G, Forsyth D, Hoiem D. Comparative Object Similarity for Improved Recognition with Few or No Examples // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. San Francisco, USA, 2010: 3525-3532 [5] Oyen D, Lane T. Leveraging Domain Knowledge in Multitask Bayesian Network Structure Learning // Proc of the 26th AAAI Conference on Artificial Intelligence. Toronto, Canada, 2012: 1091-1097 [6] Li L J, Socher R, Li F F. Towards Total Scene Understanding: Classification, Annotation and Segmentation in an Automatic Framework // Proc of the IEEE Conference on Computer Vision and Pattern Recognition. Miami, USA, 2009: 2036-2043 [7] Larlus D, Verbeek J, Jurie F. Category Level Object Segmentation by Combining Bag-of-words Models with Dirichlet Processes and Random Fields. International Journal of Computer Vision, 2010, 88(2): 238-253 [8] Fox E B, Sudderth E B, Jordan M I, et al. A Sticky HDP-HMM with Application to Speaker Diarization. The Annals of Applied Statistics, 2011, 5(2A): 1020-1056 [9] Bosch A, Zisserman A, Muoz X. Scene Classification Using a Hybrid Generative Discriminative Approach. IEEE Trans on Pattern Analysis and Machine Intelligence, 2008, 30(4): 712-727 [10] Holub A D, Welling M, Perona P. Hybrid Generative-Discriminative Visual Categorization. International Journal of Computer Vision, 2008, 77(1/2/3): 239-258 [11] Fujino A, Ueda N, Saito K. Semisupervised Learning for a Hybrid Generative/ Discriminative Classifier Based on the Maximum Entropy Principle. IEEE Trans on Pattern Analysis and Machine Intelligence, 2008, 30(3): 424-437 [12] Chapelle O, Sch lkopf B, Zien A. Semi-Supervised Learning. Cambridge, USA: MIT Press, 2006 [13] Kalal Z, Mikolajczyk K, Matas J. Tracking-Learning-Detection. IEEE Trans on Pattern Analysis and Machine Intelligence, 2012, 34(7): 1409-1422 [14] Gao J, Xie Z, Zhang J, et al. Image Semantic Analysis and Understanding: A Review. Pattern Recognition and Artificial Intelligence, 2010, 23(2): 191-202 (in Chinese) (高 隽,谢 昭,张 骏,等.图像语义分析与理解综述.模式识别与人工智能, 2010, 23(2): 191-202) [15] Xun G, Wang H F. The Development of Topic Models in Natural Language Processing. Chinese Journal of Computers, 2011, 34(8): 1423-1436 (in Chinese) (徐 戈,王厚峰.自然语言处理中主题模型的发展.计算机学报, 2011, 34(8): 1423-1436) [16] Sudderth E B, Torralba A, Freeman W T, et al. Describing Visual Scenes Using Transformed Objects and Parts. International Journal of Computer Vision, 2008, 77(1/2/3): 291-330 [17] Sudderth E B, Torralba A, Freeman W T, et al. Learning Hierarchical Models of Scenes, Objects, and Parts // Proc of the 10th IEEE International Conference on Computer Vision. Beijing, China, 2005, II: 1331-1338 [18] Blei D M. Introduction to Probabilistic Topic Models. Communications of the ACM, 2012, 55(4): 77-84 [19] Lowe D G. Distinctive Image Features from Scale-invariant Keypoints. International Journal of Computer Vision, 2004, 60(2): 91-110 [20] Oliva A, Torralba A. Building the Gist of a Scene: The Role of Global Image Features in Recognition. Progress in Brain Research, 2006, 155: 23-36
2015, Vol. 28 
