Involved people : Michel Couprie, Jean-Marie Mouchel (CERGRENE, Ecole Nationale des Ponts et Chaussées).
Associated students: Sebastien Colas, Pierre Yvonnet, Fabien Hernandez
Various products for marking bacteria can be added to solutions containing bacteria (natural water, food industrys ...). Certain products are even specific of certain types of bacteria. Once the bacteria marked, they can be observed and counted under a microscope (fig. 16). The procedure is time consuming and tiring for the eyes. Moreover, it is very difficult to evaluate the size of the observed bacteria.
Our goal is here to test and improve a robust algorithm intended to recognize the bacteria on such images. The algorithm must be able to be applied automatically to all kinds of images without the operator having to introduce different parameters for each image. Indeed, the images can be more or less encumbered by traces of particles which are not bacteria and which should be eliminated. It also happens that bacteria are included in a gel: in such a case the traditional procedures of image processing, based on a thresholding, perform very badly.
We developed an automatic data processing sequence for this type of images. One of the objectives was to check the application of our algorithms to a sufficient number of images to be representative of all the situations. A group of students of the ESIEE was associated with this development within the framework of an internal project.
