Semantic Image Clustering Using Object Relation Network [chapter]

Na Chen, Viktor K. Prasanna
2012 Lecture Notes in Computer Science  
This paper presents a novel method to organize a collection of images into a hierarchy of clusters based on image semantics. Given a group of raw images with no metadata as input, our method describes the semantics of each image with a bag-of-semantics model (i.e., a set of meaningful descriptors), which is derived from the image's Object Relation Network [5] -an expressive graph model representing rich semantics for image objects and their relations. We adopt the class hierarchies in a guide
more » ... tology as different levels of lenses to view the bag-of-semantics models. Image clusters are automatically extracted by grouping images with the same bag-of-semantics viewed through a certain lens. With a series of coarse-to-fine lenses, images are clustered in a top-down hierarchical manner. In addition, given that users can have different perspectives regarding how images should be clustered, our method allows each user to control the clustering process while browsing, and thus dynamically adjusts the clustering result according to the user's preferences.
doi:10.1007/978-3-642-34263-9_8 fatcat:mj5iad73xnbbfksij32arhxvzy