Active Image Clustering with Pairwise Constraints

People

Abstract

We propose a method of clustering images that combines algorithmic and human input. An algorithm provides us with pairwise image similarities. We then actively obtain selected, more accurate pairwise similarities from humans. A novel method is developed to choose the most useful pairs to show a person, obtaining constraints that improve clus- tering. In a clustering assignment elements in each data pair are either in the same cluster or in different clusters. We simulate inverting these pairwise relations and see how that affects the overall clustering. We choose a pair that maximizes the expected change in the clustering. The pro- posed algorithm has high time complexity, so we also pro- pose a version of this algorithm that is much faster and exactly replicates our original algorithm. We further improve run-time by adding heuristics, and show that these do not significantly impact the effectiveness of our method. We have run experiments in two different domains, namely leaf images and face images, and show that clustering perfor- mance can be improved significantly

Papers

Arijit Biswas, David Jacobs.
Active Image Clustering with Pairwise Constraints from Humans
In International Journal of Computer Vision (IJCV), 2014.
PDF 

Arijit Biswas, David Jacobs.
Active Image Clustering: Seeking Constraints from Humans to Complement Algorithms
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2012.
PDF bibtex poster

Arijit Biswas, David Jacobs.
Large Scale Image Clustering with Active Pairwise Constraints
In Combining Learning Strategies to Reduce Label Cost Workshop, International Conference in Machine Learning (ICML), 2011.
PDF
 bibtex poster

Data Sets

Leaf Dataset
Face Dataset

Code

FAST-Active-HACC-H1 (Active Image Clustering Code, Biswas, Jacobs, CVPR, 2012)

Some Results

Results