Real-time automation of leaf image segmentation is a difficult task when there are similar leaves in the background, particularly in leaf images captured in the cultivation fields. These leaf images play a key role in monitoring the growth and health of the plants. An hierarchical approach based on Kernel Linear Discriminant Analysis (KLDA) and Gaussian process regression is proposed in this paper for automating the segmentation process. In the first level, KLDA is used to discriminate the target leaf from its similar leaf background in two steps - (i) detecting the surfaces of the target leaf and (ii) detecting the edges of the target leaf. The resulting coarsely segmented image is further subjected to the second level consisting of the edge detection and morphological operations necessary for obtaining the fine segmented image. To fully automate the segmentation process, it is proposed to use the Gaussian process based regression technique for estimating the tuning parameters required for morphological processing. The proposed method is tested on a Sunflower leaf dataset and the ImageCLEF (Pl@ntleaves) dataset. The experimental results reveal the potential of the proposed method in automating the leaf segmentation process.

当背景中有相似的叶子时，尤其是在耕地中捕获的叶子图像中，叶子图像分割的实时自动化是一项艰巨的任务。这些叶片图像在监控植物的生长和健康中起着关键作用。提出了一种基于核线性判别分析（KLDA）和高斯过程回归的分层方法，以实现分割过程的自动化。在第一级中，KLDA用于在两个步骤中将目标叶片与其相似的叶片背景区分开：（i）检测目标叶片的表面和（ii）检测目标叶片的边缘。将所得的粗略分割的图像进一步进行第二级处理，该第二级包括边缘检测和获得精细分割的图像所需的形态学操作。为了使分割过程完全自动化，建议使用基于高斯过程的回归技术来估计形态处理所需的调整参数。在向日葵叶数据集和ImageCLEF（Pl @ ntleaves）数据集上测试了所提出的方法。实验结果揭示了该方法在自动分割叶片过程中的潜力。