Object recognition using graph spectral invariants

Graph structures have been proved important in high level-vision since they can be used to represent structural and relational arrangements ofobjects in a scene. One ofthe problems that arises in the analysis ofstructural abstractions of object is graph clustering. In this paper, we explore howpermutation invariants computed from the trace of the heat kernel can be used to characterize graphs for the purposes ofmeasuring similarity and clustering. We explore three different approaches to characterize the heat kernel trace as afunction oftime. These are the heat kernel trace moments, heat content invariants and symmetric polynomials with Laplacian eigenvalues as inputs. Experiments on the COIL 100 and 256 Caltech databases reveal that the proposed invariants are effective and outperform the tradition methods.