棉花苗叶片复杂背景图像的联合分割算法

发布时间：2018-11-26 19:16

【摘要】：为了实现自然光条件下棉花叶片的自动、普适和精确分割,提出一种基于马尔可夫随机场最优化模型的联合分割算法.首先用非监督的共同显著性检测算法为一组棉花苗叶片图像中每幅图像生成共同显著性图,这些显著性图被用来构造马尔科夫随机场中的内部图像能量函数;然后采用混合高斯模型对该组图像全部显著性图的共同目标(叶片)与单幅图像中叶片的差异进行建模,将其作为马尔可夫随机场最优化模型的一个新的全局约束去构造全局能量项;最后通过标准图割算法(Grabcut)和迭代使得能量函数最小化,以实现对棉花苗叶片图像的分割.按不同天气条件和不同背景拍摄600幅棉花苗叶片图像构建了数据库,在该库上的实验结果表明,该算法对于晴天、阴天和雨后图像中目标的平均正确分割率达到84.8%,87.7%和91.6%,比经典的Grabcut分别提高了10.7%,3%和10%.
[Abstract]:In order to realize automatic, universal and accurate segmentation of cotton leaves under natural light, a joint segmentation algorithm based on Markov random field optimization model is proposed. Firstly, the unsupervised joint salience detection algorithm is used to generate a common salience map for each image in a group of cotton seedling leaf images, which are used to construct the internal image energy function in Markov random field. Then the mixed Gao Si model is used to model the differences between the common target (leaf) of all salient images and the leaves in a single image. It is used as a new global constraint of Markov random field optimization model to construct global energy term. Finally, the energy function is minimized by (Grabcut) and iteration to achieve the segmentation of cotton seedling leaf image. The database was constructed by taking 600 cotton seedling leaf images under different weather conditions and different backgrounds. The experimental results on the database show that the average correct segmentation rate of the target in the images of sunny, cloudy and rain days is 84.8%. 87.7% and 91.6%, 10.7% and 10% higher than the classical Grabcut, respectively.
【作者单位】：甘肃农业大学机电工程学院;中国农业科学院农业信息研究所;
【基金】：国家自然科学基金(31501229,61461005)
【分类号】：TP391.41

【相似文献】