RSSVM-based Multi-Instance Learning for Image Categorization
Focusing on the problem of natural image categorization, a novel multi-instance learning (MIL) algorithm based on rough set (RS) attribute reduction and support vector machine (SVM) is proposed. This algorithm regards each image as a bag, and low-level visual features of the segmented regions as instances. Firstly, a collection of \visual-words\ is generated by Gaussian mixture model (GMM) clustering method, then based on the fuzzy membership function between instance and visual-word, a fuzzy histogram is computed to represent bag. As a result, every bag is transform into a single sample, which converts MIL problem to a standard supervised learning problem. Finally, RS method is used to reduce the redundant features in the fuzzy histogram, and then standard SVM classifiers are trained for image categorization. Experimental results on the COREL image set show that this algorithm is robust, and the performance is superior to other key existing MIL algorithms.
Keywords: Multi-instance learning; Image categorization; Attribute reduction; Support vector machine.
Download Full-Text
ABOUT THE AUTHOR
Daxiang Li
Daxiang LI received the BSc in electrical engineering in 1997, the MSc in 2005 and PhD in 2011 from Northwestern University (xi¡¯an China) in computer software and theory. He is now a lecturer at the school of telecommunication and Information engineering, Xi\'an University of Posts and Telecommunications. His areas of interest are image processing, machine learning, image semantic analysis and content based image retrieval (CBIR).
Daxiang Li
Daxiang LI received the BSc in electrical engineering in 1997, the MSc in 2005 and PhD in 2011 from Northwestern University (xi¡¯an China) in computer software and theory. He is now a lecturer at the school of telecommunication and Information engineering, Xi\'an University of Posts and Telecommunications. His areas of interest are image processing, machine learning, image semantic analysis and content based image retrieval (CBIR).
