Use of the angle information in the wavelet transform maxima for image de-noising

In this article, a new method of de-noising is proposed, based on the wavelet maxima. The originality of this method is in the use of the gradient angle in a multi-scale framework as the discriminatory parameter. In order to use to the best advantage the angle information, the multi-scale gradient decomposition scheme proposed by Mallat is modified thus enabling a computation of uncorrelated partial derivatives. From this computation, a selection method of multi-scale contours is put forward, having a lesser algorithmic complexity than processings based on the gradient norm. The performance of this new algorithm is illustrated using simulated data and angiography images. (C) 2000 Elsevier Science B.V. All rights reserved.