Identification d'espèces végétales par une description géométrique locale d'images de feuilles

Abstract : Plant species identification, usually performed by specialists, is based on the observation of its organs and mostly on visual criteria. Thanks to its ease of acquisition, the leaf is the most used organ. In addition, it contains important information on the taxonomy of the plant. This enables the use of computer vision and pattern recognition techniques for developing an automatic recognition process of the plant species from a leaf image. We introduce a new approach to identify plant species, based on the description of the following leaf characteristics : its shape, its salient points and its venation. First, the shape of the leaf is represented by local descriptors associated to a set of points sampled on the contour. Different multi-scale triangular representations are introduced and compared. To describe the salient points of the leaf, we propose a shape context based representation. Finally, the venation is extracted by detecting elementary linear structures with morphological tools. The venation network is described by its main directions and its spatial distribution in the context of the leaf boundary. A local matching method is used for all descriptors. Evaluations, conducted on six publicly available plant identification benchmarks, show that our approaches identify the plant species of the leaf in most of the cases and that the late fusion of the proposed descriptors improves the identification process.
Complete list of metadatas

https://hal.inria.fr/tel-01223709
Contributor : Anne Verroust-Blondet <>
Submitted on : Thursday, December 17, 2015 - 9:24:41 AM
Last modification on : Friday, May 25, 2018 - 12:02:07 PM

Identifiers

  • HAL Id : tel-01223709, version 2

Collections

Citation

Sofiene Mouine. Identification d'espèces végétales par une description géométrique locale d'images de feuilles. Traitement du signal et de l'image [eess.SP]. Télécom ParisTech, 2015. Français. ⟨NNT : 2015ENST0016⟩. ⟨tel-01223709v2⟩

Share

Metrics

Record views

572

Files downloads

392