Abstract:
Recognition of distorted images is very important in pattern recognition and computer vision, and finding efficient features which are invariant under some geometrical transforms is the key to solving this problem. The paper concentrates on deriving a novel kind of features, namely invariants based on mass/area, which are invariant with regard to translation, rotation, scale changes and reflection. A set of circles centering on the object centroid are generated first, then for each circle, the ratio of mass (or area) within the circle to the total mass (or area) is calculated, these ratios are invariants based on mass/area. Numerical experiments are carried out and show a very high recognition ratio.