CAVIAR Invariant Ridge Points

INRIA developed a new image description called a "Natural Interest Ridge", which is scale invariant descriptor of elongated shape, based on maxima in a Laplacian Scale Space.

The human form can often be represented by three dominant ridges based on the spine and 2 legs. This ridge configuration is useful for evaluating whether a detected target region contains a person or is a noise region.

Several examples of Natural Interest Ridges are:

A paper that describes the process is:

  1. Thi Thanh Hai Tran, Augustin Lux, "A Method for Ridge Extraction", ACCV '04, Asian Conference on Computer Vision, July 2004.

Back to CAVIAR home page.