Classification of indirect immunofluorescence images using thresholded local binary count features

Abstract

Computer aided classification of HEp-2 cell based indirect immunofluorescence (IIF) images is a recommended procedure for standardising autoimmune disease diagnostics. In this work a novel feature, the thresholded local binary count (TLBC) has been proposed to classify IIF images into one among six classes. The TLBC is rotational invariant and is insensitive to pixel quantization noise. It characterizes the local binary gray scale pixel information in an image. The proposed feature along with global features such as area, entropy, illumination level and mean intensity, when classified using a support vector machine gave an accuracy of 86%. This feature could help in improving the diagnostics of autoimmune diseases which is highly clinically significant.