Teeth segmentation on dental panoramic radiographs using decimation-free directional filter bank thresholding and multistage adaptive thresholding

Abstract

Dental utilization typically associated with tooth shape features which are extracted from dental panoramic radiograph image. However, because dental panoramic radiograph images usually have low contrast, we need a segmentation method that can work well on low contrast images and make the tooth shape is evident. In this paper, we propose a system to do teeth segmentation using Decimation-Free Directional Filter Bank Thresholding (DDFBT) and Multistage Adaptive Thresholding (MAT). The system is built with three main steps, which are formation of vertical and horizontal directional images using DDFBT, enhancement on directional images for teeth edge reinforcement and noise removal, and segmentation using MAT with Sauvola Local Thresholding. The experimental result on 40 teeth images shows that this system has a better performance than Otsu Thresholding, Sauvola Local Thresholding, and MAT with Niblack Local Thresholding with misclassification error (ME) and relative foreground area error (RAE) values are 17.0% and 9.7%.