An maize leaf segmentation algorithm based on image repairing technology

Abstract

Maize is one of the three cereal crops in the world. The growth status of the maize is often observed via camera. Segmentation of the maize plant from digital images is the basis of the plant phenotype. However, the traditional image segmentation algorithms are not suitable for maize segmentation due to under-segmentation or over-segmentation problems when the maize images are affected by light condition or complex background. In order to obtain an accurate maize leaf segmentation result, an automatic maize leaf segmentation algorithm is proposed in this work. The algorithm first trained two traditional maize leaf segmentation models using color and spatial features of the pixels, and then applied an image repairing technology based on spatial structure analysis of the traditional maize segmentation results, including broken points detection and matching, as well as Bezier curve fitting of broken leaves. The evaluation experiment results show that the proposed algorithm is able to segment maize plant from images that have complex background or under different light conditions. The quantitative comparative experiment shows that the proposed algorithm performs better than the traditional image segmentation algorithms. It produced segmentation results that have a similar degree of 90.23% with the ground truth images, and have a PSNR of 24.23 dB, which are both higher than those of the traditional algorithms.