Common Visual Pattern Recognition Using Hierarchical Clustering Methods with Asymmetric Networks

In this paper we propose a new method for common visual pattern identication via Directed Graphs. For this we match common feature points between two images and then apply hierarchical clustering methods to one of them to discriminate between dierent visual patterns. In order to achieve this last task we introduce a technique to obtain an asymmetric dissimilarity function AX(x;x 0 ) between the nodes X of the network N = (X;AX). For each node, the method weighs the distance between each node and the distance with all the other neighbours. A dendrogram is later obtained as the output of the hierarchical clustering method. Finally we show a criteria to select one of the multiple partitions that conform the dendrogram.