HEp-2 Cell Classification with Heterogeneous Classes-Processes Based on K-Nearest Neighbours

We present a scheme for the feature extraction and classification of the fluorescence staining patterns of HEp-2 cells in IIF images. We propose a set of complementary processes specific to each class of patterns to search. Our set of processes consists of preprocessing, features extraction and classification. The choice of methods, features and parameters was performed automatically, using the Mean Class Accuracy (MCA) as a figure of merit. We extract a large number (108) of features able to fully characterize the staining pattern of HEp-2 cells. We propose a classification approach based on two steps: the first step follows the one-against-all (OAA) scheme, while the second step follows the one-against-one (OAO) scheme. To do this, we needed to implement 21 KNN classifiers: 6 OAA and 15 OAO. Leave-one-out image cross validation method was used for the evaluation of the results.