Robust plant segmentation of color images based on image contrast optimization

• A contrast-optimization approach was proposed for plant segmentation of color images. Contrast-enhanced images were compared with index using five image datasets. The method consistently enhanced contrast and accuracy. None nine common indices robust enough to varying conditions. Plant is a crucial task in computer vision applications identification/classification quantification phenotypic features. Robust plants challenged by variety factors such as unstructured background, variable illumination, biological variations, weak plant-background contrast. Existing that are empirically developed specific may not adapt robustly imaging This study proposes new robust, automatic from background (red-green-blue, RGB) consists unconstrained optimization linear combination RGB component enhance the between regions, followed thresholding contrast-enhanced ( CEI s). validity this demonstrated datasets acquired under different field or indoor conditions, total 329 well ground-truth masks. s along 10 evaluated terms s, based on maximized foreground-background separability, achieved consistent, substantial improvements over images, an average accuracy F1 = 95%, which 4% better than best obtained indices. found sensitive conditions none them performed across straightforward, easy implement can be potentially extended nonlinear forms combinations other spaces generally useful analysis precision agriculture phenotyping.