Fig. 2

Flow of our method. In this study, a new descriptor, Segmental HOG, was developed to detect glomeruli in highly magnified microscopic images. SVM is combined with Segmental HOG to classify candidate glomeruli that passed the pre-screening stage. To do this, our detection algorithm consists of three stages. (a) In the pre-screening stage, candidate glomeruli are searched in the entire microscopic image. (b) In the segmentation stage, the boundaries of each candidate glomerulus are estimated and Segmental HOG is based on the estimated boundaries. In the classification stage, feature vectors are extracted based on the estimated segmentations, and SVM is applied to judge whether each candidate glomeruli is positive or negative