Abstract

During the development of a basic library on the parallel architecture board PC-OEIL, using "Data Flow" processors (Ramon, 87), a special customer requirement appeared. Because fast image workstations already exist, processing only binary images, most image programmers, using forced binarization, prompted us to develop a granulometry unit.

Besides the different shape parameters (area, perimeter, Feret diameter, fractal dimension approximation of contour according to Chermant & Coster method, etc...) (Chermant, 89), the primary tool for such an individual analysis is the recovery of each object. The aim of this paper is the description of the particular tips implemented around the classical morphologic method of conditional reconstruction from the first point of the object. Other methods may be as efficient (contour extraction...), but we decided to concentrate upon the morphological reconstruction.

We would like to point out the fact that improvements presented in this program are not especially dedicated to "Data Flow" architecture, but on the contrary, classical high level language programming will take fool advantage of them. Saving in execution time from several hundred to one most of the time, up to 512 to one, referred to the Basic algorithm, makes this program an original and efficient tool for automatic individual analysis. As a matter of fact, for some objects, two raster scannings are sufficient for complete reconstruction, independently of their size.

Pour citer cet article