Enhancement of ARG object recognition method

We address the problem of object recognition in computer vision. The Attributed Relational Graph (ARG) method proposed in [?] is modified to extend its applicability under object scaling condition. This consists in using representation stage, unary and binary attributes which do not involve measuring the shape of regions. The experimental result shows the new method can cope with object scaling better. In the new method we also modify the probabilistic relaxation labelling method used in[?] in order to minimise its sensitivity to the number of spurious nodes in the contextual neighbourhood.