A region based seed detection for root detection in minirhizotron images

Abstract

Automatic analysis of plant root images is very important for the analysis of root architecture and morphology as requested by ecological, agronomic and biological studies. Extending the approach in [1] we improve the detection of special root parts called seeds which are expanded to complete roots in a subsequent step. Replacing the original contour based seed detection by a region based one allows us to take more global information into account. As the seed detection influences the overall performance strongly the such acquired improvements benefit the whole root detection significantly. © Springer-Verlag Berlin Heidelberg 2003.