Attributed relational graph matching by neural-gas networks

In the past, the neural-gas (NG) network has been commonly used for clustering, classification and vector quantization of feature vectors. In this paper, a modified NG network is used to perform pattern recognition by matching attributed relational graphs. The ARG matching is formulated as an optimisation problem and the modified NG network is applied to solve it. As every scene vertex is matched to the best matching model vertex, there are some spurious matches in the mapping generated by the NG network. A pose clustering algorithm is used to eliminate these spurious mappings and to estimate the pose parameters. We present experimental results to demonstrate the proposed procedure