A Webber local binary pattern descriptor for pancreas endoscopic ultrasound image classification

Abstract

A Webber local binary pattern (LBP) descriptor (WLBP) is proposed to improve the classification performance between chronic pancreatitis and autoimmune pancreatitis from endoscopic ultrasound images. First, the differential excitation map and orientation map are extracted according to Webber local descriptor from each image. Then an orientation weight correction mechanism based on the orientation map is used to adjust the gray value of neighbors in differential excitation map. Finally the corrected differential excitation map is used to establish the LBP histogram and obtain a 10- dimensional feature vector. The proposed feature is evaluated on pancreas endoscopic ultrasound images collected by Gastroenterology Department of Changhai Hospital from 2005 to 2011. The result suggests that the WLBP shows better classification capacity than the LBP, the Webber local descriptor (WLD) and other traditional features on pancreas endoscopic ultrasound images and may provide valuable reference for clinical differential diagnosis.