A PCA-like rule for pattern classification based on attributed graph

Attributed graph (AG) is a useful data structure for representing a complex pattern, However, the existing methods for image understanding based on this structure all encounter the problem of attributed graph matching (AGM) which is usually a hard combinatorial problem with very high computational complexity. This paper suggests to separate the AG-based image understanding into two steps-classification and correspondences building. A principal component analysis (PCA) like rule is proposed for pattern classification based AG without involving the hard combinatorial problem of AGM.