Shape Analysis

Shape plays an important part in the processing of visual information, from art through to science, and within computer vision there have been many applications of shape to aid in the analysis of images.

We have developed shape descriptors for a variety of types of shape, for instance to measure linearity, circularity, ellipticity, rectangularity, and triangularity (for more detail on convexity measures see here).

There are no standard methods for computing rectilinearity, but it has many potential applications. Rectilinear structures often correspond to human-made structure, and are therefore justified as attentional cues for further processing. For instance, in aerial image processing and reconstruction, where building footprints are often rectilinear on the local ground plane, building structures, once recognized as rectilinear can be matched to corresponding shapes in other views for stereo reconstruction.

The images below show a Digital Elevation Model Some simple noise filtering and segmentation techniques were applied to produce a set of polygons, and they are coloured with intensities proportional to their rectilinearity; thus rectilinear shapes generally appear bright. The third image shows the effects of applying a snake-based refinement which incorporates rectilinearity in addition to proximity to the original data. Deviations from rectilinearity have been corrected if it does not require excessive deformation of the shape.

More details are given in:

return to Paul Rosin's homepage