Medial-point description of shape: a representation for action coding and its psychophysical correlates

We describe a region-based shape representation that might be particularly useful from a biological perspective because it promotes the localization of objects, and object parts relative to each other. The proposed medial-point representation is similar to medial-axis type representations, but it is more compact. The medial points are those points along the medial axis that are equidistant from the longest segments of the boundary, therefore they represent the largest amount of edge information. The main advantage is that the original image can be reduced to a small number of points. We also provide psychophysical correlates of the representation for shapes with increasing complexity. Using a reverse mapping technique, we find that variations of contrast sensitivity within figures are defined by the shape of the bounding contour, and the peaks in the sensitivity maps correspond to the medial points of the proposed representation.