Visual clustering with minimax feature fusion

To leverage multiple feature types for visual data analytics, various methods have been presented in Chaps. 2 – 4. However, all of them require the extra information, e.g., the spatial context information and the data label information. It is often difficult to obtain such information in practice. Thus, pure multi-feature fusion becomes critical, where we are given nothing but the multi-view features of data. In this chapter, we study multi-feature clustering and propose a minimax formulation to reach a consensus clustering. Using the proposed method, we can find a universal feature embedding, which not only fits each feature view well, but also unifies different views by minimizing the pairwise disagreement between any two of them. The experiments with real image and video data show the advantages of the proposed multi-feature clustering method when compared with existing methods.