Feature co-occurrence for visual labeling

Due to the difficulties in obtaining labeled visual data, there has been an increasing interest to label a limited amount of data and then propagate the initial labels to a large amount of unlabeled data. In this chapter, we propose a transductive label propagation algorithm by leveraging the advantages of feature co-occurrence patterns in visual disambiguity. We formulate the label propagation problem by introducing a smooth regularization that ensures similar feature co-occurrence patterns share the same label. To optimize our objective function, we propose an alternating method to decouple feature co-occurrence pattern discovery and transductive label propagation. The effectiveness of the proposed method is validated by both synthetic and real image data.