Encoding rotation invariant features in HEP-2 cell classification

Detection of the antinuclear antibodies (ANAs) in patient serum can be accomplished via indirect immunofluorescence (IIF) technique in laboratory. However, the currently practiced manual detection system relies on specialists to observe HEp-2 slides, which suffers from severe drawbacks like being subjective and time-consuming. In this study, we present an automatic classification system for staining patterns of HEp-2 cells. We propose to extract two kinds of rotation invariant descriptors, i.e., Pairwise Local Ternary Patterns with Spatial Rotation Invariant (PLTP-SRI) feature and Bag-of-Words (BoW) representation based on dense SIFT. The descriptors are concatenated to achieve the advantages of both highly descriptive ability and robustness against rotations. Incorporated with a linear Support Vector Machine (SVM) classifier, our proposed algorithm demonstrates its effectiveness by testing on two datasets: ICPR12 dataset and ICIP2013 training dataset.