Hierarchical classification of HEP-2 cell images using class-specific features

The paper proposes a class-specific feature assisted automatic classification approach of microscopic HEp-2 cell images. Unlike traditional methods our method highlights two important aspects: (1) the visual characteristics of classes to formulate class-specific image features and (2) the classification task is treated as hierarchical verification sub-tasks. Thus, the overall classification problem is modeled as a verification of each class, using its class-specific features. We have demonstrated that the proposed method yields a high classification rate utilizing simple and efficient features with only (20%) of the data for training. Additionally, we also experimentally analyze the crucial aspects, such as comparison with a traditional non-hierarchical framework and performance evaluation on low contrast images which is useful for early disease detection.