Biologically-Inspired Dense Local Descriptor for Indirect Immunofluorescence Image Classification

Abstract

This work deals with the design of a classification method for cells extracted from Indirect Immunofluorescence images. In particular, we propose to use a dense local descriptor invariant both to scale changes and to rotations in order to classify the six categories of staining patterns of the cells. The descriptor is able to give a compact and discriminative representation and combines a log-polar sampling with spatially-varying gaussian smoothing applied on the gradients images in specific directions. Bag of Words is finally used to perform classification and experimental results show very good performance.