Landmark detection on cephalometric radiology images through combining classifiers

This paper, presents a new cephalometric landmark localization method based on combining two classifier results. Initially, a classifier based on histograms of oriented gradients makes a first estimation of the potential windows, and then a second classifier, based on histograms of gray profile, classifies the detected windows. By combining the results of these two classifiers, final decision is made about the landmark window location. HOG features gather edge profiles in the image and making decision for the most proper window in some detection windows needs more information than just edge profiles. By adding the gray profile features to the system and combining the results in a proper manner, detection performance increases significantly for a range of hard to easy landmarks.