A parameter selection method for wavelet shrinkage denoising (Q1431667): Difference between revisions

The authors suggest an \(\| f-g_\lambda \|_{L_2}\) versus \(\| g_\lambda \|_{B^1_{1,1}}\) \(L\)-curve criterion which they call \(H\)-curve criterion for the choice of the threshold parameter \(\lambda\) in soft wavelet threshold denoising of images. Here \(f\) denotes the noisy image and \(g_\lambda\) its denoised version with threshold \(\lambda\). They show by some experiments that their criterion works better than the classical Donoho choice and the generalized cross validation.