HEp-2 Cells Classification Using Morphological Features and a Bundle of Local Gradient Descriptors

Abstract

A system for automatic classification of staining patterns in IIF imaging is presented. A full pipeline of pre-processing, feature extraction and classification stages is designed in order to overcome specific challenges posed by the nature of the data. In the preprocessing stage the images are subjected to normalization and de-noising using a sparse representation-based technique. A set morphological features, extracted using multi-level thresholding, is combined with a bundle of local gradient descriptors, selected so as to encode textural and structural information of the fluorescent patterns in multiple scales. The proposed method was evaluated using a dataset with over 10K images achieving over 90 percent of classification accuracy.