Persistent Betti Numbers for a Noise Tolerant Shape-Based Approach to Image Retrieval

In content-based image retrieval a major problem is the presence of noisy shapes. It is well known that persistent Betti numbers area shape descriptor that admits a dissimilarity distance, the matchingdistance, stable under continuous shape deformations. In this paper wefocus on the problem of dealing with noise that changes the topologyof the studied objects. We present a general method to turn persistentBetti numbers into stable descriptors also in the presence of topologicalchanges. Retrieval tests on the Kimia-99 database show the effectivenessof the method.