An hierarchical approach for automatic segmentation of leaf images with similar background using kernel smoothing based Gaussian process regression

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.