||A binary image holds no information on the sub-pixel position of the edge of a sampled object since each sample is considered to be either fully inside or fully outside the object. Therefore, distance values for traditional distance transforms are constrained to values attained in an integer grid. In this presentation, a model where the intensity values represent the area coverage (the fraction of the pixel that meet the object) is used to approximate the position of the object contour. We assume that the area sampled object is binary and has a smooth contour. When the assumptions are met, we get a distance transform with higher accuracy compared to traditional methods.
This project has been carried out in collaboration with Stefan Gustavson, Media and Information Technology, Linköping University.