基于G-MRF模型的玉米叶斑病害图像的分割

【目的】图像分割是作物病害自动识别系统实现的难点之一,前人研究大多采用基于阈值或聚类的分割算法,方法简单、易于实现,但分割精度较低。本文引入高斯模型的Markov随机场分割模型(G-MRF),对玉米叶部病斑图像进行分割试验,以期提高分割精度。【方法】在VC6.0下实现了G-MRF分割模型,G-MRF既利用了图像像素的灰度信息,又通过像素类别标记的Gibbs光滑先验概率引入了图像的空间信息,是能较好地分割含有噪声图像的算法。采用该算法对大斑病、小斑病、灰斑病和弯孢菌叶斑病等4种主要玉米叶部病害的图像进行了分割测试,并与基于阈值和基于Gauss模型的分割算法进行比较。【结果】基于G-MRF分割模型的分割,目标区域的一致性和边缘的清晰方面明显好于基于阈值和Gauss模型的分割算法,其平均正确分类率达96.35%,分别较基于阈值和基于Gauss模型的分割算法高出3.75%和4.03%,差异达到显著水平。【结论】基于G-MRF模型的分割算法鲁棒性高,能够有效地将病斑区域从叶片部分离,分割正确分类率达96.35%,可用于玉米叶斑类病害图像的分割。

Maize Leaf Disease Spots Segmentation Based on Gauss-MRF Model

Objective] Image segmentation is one of the difficult issues in crop disease automatic identification systems, in present research, algorithms based on threshold or based on clustering were mostly used. These methods are simple and easy of implementation, but the precision is low. In order to improve the segmentation precision of maize disease spots, Gauss-MRF model was introduced to segment maize disease spot images, Method] The algorithm was implemented on VC6.0. Gauss-Markov random field model takes advantage of both image intensity and spatial information imposed by Gibbs smoothness prior about the pixel labels and thus can be used to effectively segment the noised images. Four kinds of corn disease images were segmented, and segmentation results were compared with algorithms based on threshold and Gauss model. Result] The segmentation algorithm based on G-MRF model showed higher precision, and the segmentation results showed accurate and closed boundaries. The mean correct classification ratio of G-MRF model was up to 96.35% which was higher than algorithms based on threshold and Gauss model, respectively, by 3.75% and 4.03%. Conclusion] The research shows that the segmentation algorithm based on G-MRF model is satisfactory to separate disease part from normal part of leaves. The mean correct classification ratio is 96.35%. It is effective in segmenting and processing maize disease spots images.