Local Multilevel Threshold filtering in the curvelet domain
A. P. de Franco*, Fernando Sergio Moraes · 2015
Seismic reflection images are usually corrupted by different kinds of undesirable noise, which may compromise the interpretation process. Noise can be classified as coherent or incoherent. The Curvelet Transform (CT) is a relatively recent tool that brings the image to a higher dimension sparse domain with multiscale and multidirectional expansions. This transform lends itself particularly well to represent features that are smooth along a curve and have an oscillatory behavior in the normal direction, just like the main features of a seismic data. The higher sparsity promoted by CT allows that a few large coefficients represents the signal components while incoherent energy, like random noise, is spread amongst a great number of small coefficients. Additionally multidirectional decomposition turns out as a powerful feature in the analysis of seismic events, which have preferred directions. In this work is presented the development of a threshold estimate based on a windowed neighborhood Root Mean Square (RMS) that works as a weight array for curvelets coefficients at each location, and consequently, at each scale and direction.