Affine invariance contour descriptor based on iso-area normalisation

In pattern recognition, by observing a contour extracted from a shape, the position of the points of this contour can be normalised by using a set of triangles with the same area whose vertices consist of the centroid of the contour and a pair of successive points on the contour. Contour characterisation can be obtained by lowpass filtering and observing the resulting changes in the areas of the triangles. It is proved that, for a filtered contour, the areas of the triangles are linear under affine transforms. The experimental results show that the approach is resistant to affine transformations and noise