Abstract

Title : Image Retouching Using Wavelet Maxima
Author : Jacques Froment

We introduce a process to easily retouch digital images. Image retouching is a quite common practice for press or art publisher. It usually requires a time expensive work on graphics computer, where the operator has to manually change a large number of pixels. One of the complexity of the work comes from artifacts which may appear, leading to a reject of the picture as clearly touched up. Multiscale analysis and synthesis provide a way for retouching images without artifacts, by mixing the coefficients in the transform domain. Examples of humain parts as an eye put into an object, using the Laplacian pyramid, have been shown in the past. Such image fusion process has been used more recently to combine images from different sensors. But the problem of working with a large number of coefficients still remains. If we are interesting about form retouching, such as the nose shape of an actress, the operator should have only to redraw the edge of the nose. We are seeking an algorithm able to reconstruct a whole image with natural appearance, by stating only some slight geometric changes in the edges. Mallat and Zhong have shown that multiscale edges can be detected and characterized with a wavelet transform. In this article, we show how this multiscale representation can be used to retouch images by working on edges only.