Highly accurate HEp-2 Specimen Image Segmentation and Classification using LBP and HoG using CNN classifier

Abstract

Classification of Human Epithelial-2 (HEp-2) cell is noteworthy in the analysis of Autoimmune Diseases (AD). In this work, HEp-2 Specimen Image Segmentation and Classification utilizing deep convolution neural network (DCNN) with histogram of oriented gradients (HOG) and Local Binary Pattern (LBP) highlights (HEp2-SIC-DCNN). We proposed to order HEp-2 example pictures. HEp-2 cells are preprocessed to improve the classification exactness. Contrast limited adaptive histogram equalization technique (CLAHE) is used to improve the pixel quality. To extract features, HOG and LBP are considered in the pixel domain. The proposed method can classify different x classes such as homogeneous, speckled, nucleolar, centromere, nuclear membrane and Golgi. To enhance the training process data augmentation is used. Various sets of data are generated by rotating images in different angles to get more sample images. HEp2-SIC-CNN is tested with various images from ICPR 2014 database with various condition such as scale and rotation. Accuracy (MCA) of about 95.56%, which is every much higher compared to previous work related to available ICPR 2014 dataset.