Abstract Most SVM-based active learning methods are challenged by the small sample problem and the asymmetric distribution problems. A SVM-based active relevance feedback scheme is presented which deals with SVM ensemble under semi-supervised setting to augment the diversity among the individual SVM classifiers, thus a powerful ensemble classification model is obtained. Meanwhile, the powerful ensemble model is helpful to identify the most informative images for active learning. Moreover, aggregation method, termed as bias-weighting, is used within the semi-supervised ensemble framework to tackle the asymmetric distribution between positive and negative samples. Under the influence of bias-weighting, the ensemble classification model pays more attention on the positive samples than the negative ones. Experimental results validate the superiority of the presented scheme over several existing active learning methods.
[1] Datta R, Joshi D, Li Jia, et al. Image Retrieval: Ideas, Influences, and Trends of the New Age. ACM Computing Surveys, 2008, 40(2): 1-60 [2] Zhou X S, Huang T S. Relevance Feedback in Image Retrieval: A Comprehensive Review. Multimedia System, 2003, 8(6): 536-544 [3] Chen Kejia, Jiang Yuan, Zhou Zhihua. An Image Retrieval Method Based on Active Relevance Feedback. Pattern Recognition and Artificial Intelligence, 2005, 18(4): 480-485 (in Chinese) (陈可佳,姜 远,周志华.基于主动相关反馈的图像检索方法.模式识别与人工智能, 2005, 18(4): 480-485) [4] Huang T S, Dagli C K, Rajaram S, et al. Active Learning for Interactive Multimedia Retrieval. Proc of IEEE, 2008, 96(4): 648-667 [5] Tong S, Chang E. Support Vector Machine Active Learning for Image Retrieval // Proc of the 9th ACM International Conference on Multimedia. Ottawa, Canada, 2001: 107-118 [6] Jiang Wei, Er Guihua, Dai Qionghai. Boost SVM Active Learning for Content-Based Image Retrieval // Proc of the 37th Asilomar Conference on Signals, Systems and Computers. Pacific Grove, USA, 2003, Ⅱ: 1585-1589 [7] Wang Lei, Chan K L, Zhang Zhihua. Bootstrapping SVM Active Learning by Incorporating Unlabelled Images for Image Retrieval // Proc of the IEEE International Conference on Computer Vision and Pattern Recognition. Madison, USA, 2003: 629-634 [8] Hoi S C H, Jin Rong, Zhu Jianke, et al. Semi-Supervised SVM Batch Mode Active Learning and Its Applications to Image Retrieval. ACM Trans on Information Systems, 2009, 27(3): 1-29 [9] Zhou Zhihua. When Semi-Supervised Learning Meets Ensemble Learning // Proc of the 8th International Workshop on Multiple Classifier System. Reykjavik, Iceland, 2009: 529-538 [10] Wu Jun, Lin Zhengkui, Lu Mingyu. Asymmetric Semi-Supervised Boosting for SVM Active Learning in CBIR // Proc of the ACM International Conference on Image and Video Retrieval. Xian, China, 2010: 182-188 [11] Tao Dacheng, Tang Xiaoou, Li Xuelong, et al. Asymmetric Bagging and Random Subspace for Support Vector Machines-Based Relevance Feedback in Image Retrieval. IEEE Trans on Pattern Analysis and Machine Intelligence, 2006, 28(7): 1088-1099 [12] Rocchio J J. Relevance Feedback in Information Retrieval // Salton G, ed. The SMART System. New York, USA: Prentice-Hall, 1971: 313-323 [13] Zhang Ruofei, Zhang Zhongfei. BALAS: Empirical Bayesian Learning in the Relevance Feedback for Image Retrieval. Image and Vision Computing, 2006, 24(3): 211-223 [14] Chang C C, Lin C J. LIBSVM: A Library for Support Vector Machines [DB/OL]. [2010-03-06]. http://www.csie.ntu.edu.tw/~cjlin/libsvm
2010, Vol. 23 
