Abstract
In pattern recognition, graphs are usually used to represent the features of objects due to their richer representability than feature vectors. However, some disadvantages limit the development of graph pattern recognition, such as there is few classsifier that can be used on graph patterns and the crucial method of graph matching has a high computational complexity. In this paper, we focus on the structure analysis of tree graphs. By defining the propagations of probability among graph structures, we propose a kernel method for tree graphs with no graph matching. The experimental results with the tree graph representation of silhouette images confirm the effectiveness of the proposed method.
