<title>Soft image processing</title>

Abstract

Edges of general images are generally made of the noise (or singular) points of images. Some purely discrete and very simple algorithms, like Sobel one, are able to approximate these edges, but, unhappily, they lack accuracy. Using the restriction of gray level images to 3 X 3 masks interpreted as 3D digital surfaces, we give a new way of computing exactly the normal vector, or gradient, at regular points. We derive from this study a new discrete algorithm, based on Max Area Triangles contained in 3 X 3 masks, able to make a finer distinction between regular and singular point on any image and giving softer boundaries.