A new multi-label scene categorization method using color incorporated biologically inspired features and MIML SVM

作者：Huang, Shuangping^*; Jin, Lianwen

来源：ICIC Express Letters, Part B: Applications , 2011, 2(2): 287-293.

• Biologically inspired
• Biologically inspired models
• Clustering techniques
• Color histogram
• Feature dimensions
• Heterogeneous features
• MIML SVM
• Multi-label
• Scene categorization
• Scene classification
• Scene image
• Semantic categorization
• Sub-regions
• SVM classifiers

Semantic categorization of real world scene images is a challenging task due to the high level of ambiguity involved. In this paper, we propose a new method for multi-label scene classification using an improved biologically inspired feature that incorporates with color histogram and multi-instance multi-label learning (MIML) SVM classifier. In this new method, each scene image is firstly divided into a set of sub-regions, then their color-based histograms and C2 features of Biologically Inspired Model (BIM) are extracted. The two heterogeneous features are weighted concatenated as an instance representation for the scene image sub-region. Clustering technique is introduced to the BIM so that it produces not only better prototypes but also lower feature dimensions. MIML SVM classifier is used for final categorization. Experimental results indicate that our approach outperforms previous method significantly.

• 出版日期2011-4
• 单位华南农业大学; 华南理工大学