Skip to main content
Fig. 3 | Plant Methods

Fig. 3

From: A two-step registration-classification approach to automated segmentation of multimodal images for high-throughput greenhouse plant phenotyping

Fig. 3

The pipeline of data processing for the registration-classification based plant multimodal plant image segmentation. From top to bottom: color distance maps between reference (empty background) and plant containing FLU and VIS images are computed and subsequently used for co-registration of FLU and VIS images. High contrast FLU images are automatically segmented using the fixed z-score threshold for local color distance between the reference (background) and plant-containing images. Binary mask of the registered FLU image is applied to detect plant structures in the VIS image. Due to motion artefacts regions of the VIS image overlaid by registered FLU mask may contain plant as well as marginal background structures. Unsupervised k-means clustering of Eigen-colors is applied to generate a compact representation of pre-segmented VIS images by a small number (10–30) of color-region centroids. Pre-trained classification models of plant/background color-regions and small object filters are applied to differentiate between plant and non-plant structures

Back to article page

Plant Methods

ISSN: 1746-4811