Cephalometric landmarks localization based on Histograms of Oriented Gradients

In this paper, we explored the use of certain image features, block-wise histograms of local orientations. They are used in many current object recognition algorithms, for the task of locating cephalometric landmarks on X-ray images. After reviewing existing cephalometric landmark detection systems, we show experimentally that grids of Histograms of Oriented Gradients (HOG) descriptors significantly perform well on this task. The influence of bin and detection window size on performance for three landmarks has been studied. It has been shown that, fine orientation binning, large enough detection windows which contain whole the features around landmark are all important for good estimation of landmark position.