A framework for the extraction of quantitative traits from 2D images of mature Arabidopsis thaliana

• Published in: Machine vision and applications 2016-07, Vol.27 (5), p.647-661
• Main Authors: Augustin, Marco, Haxhimusa, Yll, Busch, Wolfgang, Kropatsch, Walter G.
• Format: Article
• Language: English
• Subjects: Communications Engineering; Computer Science; Continuity (mathematics); Extraction; Flattening; Image acquisition; Image Processing and Computer Vision; Image reconstruction; Machine vision; Manuals; Networks; Pattern Recognition; Segments; Special Issue Paper; Topology; Two-dimensional analysis; Vision systems
• ISSN: 0932-8092; 1432-1769
• DOI: 10.1007/s00138-015-0720-z

In this work, we propose an image-based phenotyping framework for the determination of quantitative traits from mature Arabidopsis thaliana plants. Two-dimensional (2D) images taken from the dried and flattened plants are analyzed regarding their geometry as well as their branching topology. The realistic branching architecture is hereby reconstructed from a single 2D image using a tracing approach with a semi-circular search window. The centerline segments of the tracing procedure are subsequently merged and labeled based on a hierarchical approach combining continuity properties with geometrical and topological information determined during tracing. This paper covers a detailed description of the proposed plant phenotyping pipeline from the image acquisition process until the extraction of the quantitative traits. The framework is evaluated using a set of 106 images and compared to a manual phenotyping approach as well as a semi-automatic image-based approach. The most relevant results of this evaluation are presented.