Force-based registration method using attribute values

In this paper, a force-based method for registration of a pattern with consideration of attributes (color, grey level, curvature) is described. In the method, a pattern is represented as a group of points which are sampled from the pattern. When we consider registration of a two dimensional pattern with color (red, green, blue) information, model pattern (the pattern at initial state) and aimed pattern (the pattern after some movement) are input in the same form: respectively, P_i (x_i,y_i,r_i,g_i,b_i ;i=1~n) and P´_j(x´_j,y´_j,r´_j ,g´_j,b´_j;j=1~m). The model pattern is registered to the aimed pattern by simulating the movement of the model pattern caused by attractive forces applied to each point of P_i from all points of P´_j. The magnitude of the force is defined using five dimensional space (x, y, r, g, b) so that stronger forces are exerted between “closer” points while considering similarity of their attribute values. In this paper the algorithm of the method is represented and its performance characteristics are examined. Experiments on registration of pieces of a jigsaw puzzle show that this simple method is robust and useful