Algorithmic recognition of biological objects

Citation
T. Bernier et Ja. Landry, Algorithmic recognition of biological objects, CAN AGR ENG, 42(2), 2000, pp. 101-109
Citations number
13
Language
INGLESE
art.tipo
Article
Categorie Soggetti
Agriculture/Agronomy
Journal title
CANADIAN AGRICULTURAL ENGINEERING
ISSN journal
0045-432X → ACNP
Volume
42
Issue
2
Year of publication
2000
Pages
101 - 109
Database
ISI
SICI code
0045-432X(200004/06)42:2<101:AROBO>2.0.ZU;2-V
Abstract
An algorithmic method of object recognition to identify and count fungal sp ores in microscopic digital images is presented. The development of this pr ocess is a key element and cornerstone of a large-scale research program ul timately aimed at reducing fungicide application. The program, as a whole, is an attempt to build a machine based system in order to improve the abili ty of researchers to assess the population of pathogenic fungi within agric ultural crops and thus more accurately target fungal pests. A three pass me thod was used: a preliminary pass in order to narrow the search space down to only the areas that contain spore-like darkening: a second pass that hig hlights both the center and the surrounding edge of the spore and produces a secondary image; and a third pass in which a template is matched to the s econdary image. After the final pass, the list of positions and orientation s of spores is reviewed and the conflicting and less Likely positions are e liminated. The goal of the method is to accurately count the spores in the minimum amount of time. The resulting time is between 0 and 21 s of analysi s on a 100 Mhz Pentium computer for a 64 by 64 pixel image. The algorithm, as implemented, demonstrated an accuracy of +/- 5.3% on low quality images, which is less than the assumed error of humans performing the same task an d is tolerant of partial occlusion. The system is loosely based on biologic al vision, is extremely versatile, and could be adapted for the recognition of virtually any object in a digitized image.