Analysis of edge and corner points using parabolic dictionaries (Q2300753)

It is difficult for classical wavelet analysis to effectively deal with directional characteristics in high dimensional signals. A number of directional representation dictionaries constructed by parabolic scaling law have been developed to solve the above problems, including curvelets, shearlets, contourlets and so on. These anisotropic directional dictionaries are constructed in various ways, and they are proved to be very useful for many practical and theoretical topics. The similarity in approximation properties of these parabolic directional dictionaries has been studied in discrete and continuous parabolic molecular frameworks. The adequacy of parabolic directional dictionaries in edge detection has been studied in microlocal analysis, involving dictionaries including second generation curvelets and classical band-limited shearlets, even all admissible parabolic molecules, where the focus of the research is rates of decay of frame coefficients with respect to the dilation parameter. The construction of compactly supported shearlets, and that of a family of high-pass filters whose relationship between angle and radial filters satisfy the parabolic scaling law have also been investigated. Building on a family of earlier studies mentioned above, this paper is devoted to extending and generalizing analysis for corner and edge feature detection of 2D regions in the framework of continuous parabolic molecules (CPM). In this general framework, no assumptions are posed on how these parabolic dictionaries are constructed in the frequency or in the space domain, no assumptions are posed on the supports of parabolic dictionaries, so their parabolic dictionaries satisfy some rather mild assumptions. This paper distinguishes between edge and corner points, rates of decay of frame coefficients for edge points and edge points are investigated and depicted. Moreover, it concludes that virtually any anisotropic scaling law yields entirely analogous results to the the parabolic scaling law in the context of edge and corner detection. As is pointed out in this paper, there are some drawbacks of this general framework. Their results are of only qualitative nature, that is, the dependence of the decay rates on the order of the given CPM family and on the smoothness of the boundary in the vicinity of the edge point are not made more explicit here; the upper bounds on frame coefficients are provided, yet the lower bounds on frame coefficients are not established, which are crucial to ensure proper detection of edge and corner points. This work could help us understand directional dictionaries and their applications to detection of edge and corner points in the level of generality.