Prediction of Edge Displacement due to Image Contrast

This study proposes a method to predict the amount of edge displacement (ED) caused by brightness difference in imaging. ED is a measure of the deviation of imaged edges from their true edge locations in the real world. A set of target sheets were designed to obtain reference lines and collect edge profiles for a given image contrast. A total of 330 target-sheet images of varying contrast and camera-to-object distances were used, with their EDs estimated using a Gaussian function. The EDs and pixel values of dark and light areas either side of an edge were then used as observations to predict EDs at a regular grid of pixel values using kriging, which is shown to be an effective method to fit the ED observations to a surface with high prediction accuracy. In addition, simulations indicate the need to correct EDs for accurate space intersection.