Gaussian Process Shape Models for Bayesian Segmentation of Plant Leaves

Abstract

We develop a novel probabilistic model for multi-part shapes based on Gaussian processes, which we apply to model rosette leaves of Arabidopsis plants. Our model incorporates domain knowledge of Arabidopsis leaves in two ways. First, leaves are modeled using two anatomical parts: a blade and a petiole. We model the two regions with separate Gaussian processes, with a smoothness constraint at the boundary. Second, we constrain all leaf petioles to initiate at the rosette center, which is also modeled. This Bayesian prior is combined with a simple likelihood function over foreground pixels to perform image segmentation by optimizing a posterior distribution. A simple datadriven approach is used to over-segment the image, then excess leaves are pruned using a Bayesian model selection criterion. We show that our approach is effective, even with minimal training data.

Leaf Segmentation and Tracking in Arabidopsis thaliana Combined to an Organ-Scale Plant Model for Genotypic Differentiation

“…influence leaf size (McDonald et al, 2003; Scoffoni et al, 2011), but environmental influences, in particular light, can also influence leaf shape (Tsukaya, 2005), even though the underlying mechanisms are not yet fully understood. Therefore, it is all the more important to test whether more sophisticated shape models (Herdiyeni et al, 2015; Simek and Barnard, 2015) could replace our relatively simple ellipse-based one and help estimate leaf areas more precisely or even increase the rate of leaf detection. Likewise, regarding leaf tracking, a promising approach would be to develop an iterative method in which, after a first step based on our original approach, the parameterized model thus obtained could be used instead of the sigmoid extrapolation to predict individual leaf areas at the next time step.…”

“…The segmentation problem has been approached in various ways, with recent contributions using ellipsoid leaf-shape models (Aksoy et al, 2015), Gaussian process shape models under a Bayesian approach (Simek and Barnard, 2015) or machine-learning (Pape and Klukas, 2015). The approach used here was inspired from Apelt et al (2015).…”

Edge Tracing using Gaussian Process Regression

“…Tu and Zhu [36] proposed a Bayesian statistical framework to image segmentation by learning object appearance models for different region types, placing prior distributions over region size, region number and boundary smoothness. Simek and Barnard [37] used a two-dimensional Gaussian process regression model to segment Arabidopsis leaves by modelling the blade and petiole as two random functions, joining them at their boundaries using a smoothing constraint. Our approach is in a similar vein to Simek and Barnard, but there is no trained prior based on a priori information.…”

Data Augmentation From RGB to Chlorophyll Fluorescence Imaging Application to Leaf Segmentation of Arabidopsis Thaliana From Top View Images

“…Segmentation of Arabidopsis thaliana leaves in RGB images has been highly studied since the introduction of the CVPPP challenge. If in 2014 and 2015 the contributions of this challenge proposed methods based on models [20,27,21], most of the participants have so far mainly tackled the challenge with deep neural network [29,26,31]. In this work we did not propose any innovation on this side and rather work on a standard neural network architecture but applied it for the first time on another imaging modality.…”