Figure 6.

Oriented bounding box refinement: (a) horizontal bounding box, (b) superpixel segmentation, (c) Coarse Contours representation, (d) rotating points to get potential bounding box and superpixel set box, (e) box refinement in principal orientation, and (f) final result. Notice that, in (d) the rotation operation was not for the whole image but for the points only. In (d) and (e), the enclosing bounding box of the coarse contours, superpixel set, and refined bounding box are shown in blue, white, and yellow, respectively.