Image denoising using the ridgelet bi-frame

We are concerned with the performance evaluation of the ridgelet bi-frame for image denoising application. The ridgelet bi-frame is a new (as far as we know) bi-frame system that can efficiently deal with straight singularities in two dimensions. We show that, for images dominated by straight edges, the ridgelet bi-frame can obtain much better restoration results than wavelet systems. We also investigate the statistical properties of the ridgelet bi-frame coefficients of these images. Results indicate that the marginal distribution of ridgelet bi-frame coefficients has higher kurtosis than that of wavelet coefficients of the same images. We describe a simple method through which statistical denoising algorithms previously developed in the wavelet domain can be conveniently introduced into the ridgelet bi-frame domain. In addition, we use the ridgelet bi-frame to construct another new bi-frame system referred to as the curvelet bi-frame, which can be viewed as a generalized version of the curvelet. Experiment results show that the simple hard-threshold procedure in the curvelet bi-frame domain produces restoration results comparable with those due to the state-of-the-art denoising methods.