Curvature estimation of oriented patterns

In digital pictures curvature which characterizes the local geometry of features cannot be calculated exactly and can only be estimated. This paper presents two methods to estimate curvature of oriented patterns at different scales which can be used even if the structures are very close. The first approach assumes that a structure is locally defined by an implicit isointensity contour. Curvature is obtained by a direct computation stemming from the differerential geometry. The second approach is based on an explicit representation of a structure defined by a set of points initially extracted. We assume this set represents an osculating circle and fit it with a circular arc. Both methods are applied to curvature estimation of seismic images for geological interpretation.