We study the inverse problem of recovering the scatterer shape from the far-field pattern(FFP) of the scattered wave in presence of noise. This problem is ill-posed and is usually addressed via regularization. Instead, a direct approach to denoise the FFP using wavelet technique is proposed by us. We are interested in methods that deal with the scatterer of the general shape which may be described by a finite number of parameters. To study the effectiveness of the technique we concentrate on simple bodies such as ellipses, where the analytic solution to the forward scattering problem is known. The shape parameters are found based on a least-square error estimator. Two cases with the FFP corrupted by Gaussian noise and/or computational error from a finite element method are considered. We also consider the case where only partial data is known in the far field.