Understanding More Types of Social Relationships Using Clothing and Distance Metric Learning

Image-based social relationships classification is an emerging and challenging problem in social media analysis. Practically, social relationships can provide intelligent entities with a deeper understanding of human behavior and emotion. Essentially, as an extension of our former study, we proceed with the following work. First, an extended social relationships database named SRI2.0 is introduced based on the coverage of social relationships categories. Second, we propose to exploit clothing information to extract the relevance of clothing characteristics among the group in images. Finally, we propose a metric learning method by using the correlation similarity measure to better highlight the differences between images with different types of social relationships. Experimental results demonstrate the efficiency of our proposed methods.