基于多模态竞争水平集的复杂背景下植物叶片图像分割方法研究

项目名称： 基于多模态竞争水平集的复杂背景下植物叶片图像分割方法研究

项目编号： No.61472280

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 张晓平

作者单位： 同济大学

项目金额： 84万元

中文摘要： 在植物叶片数字化研究中，对图像中叶片分割效果的好坏直接影响到叶片信息的提取和最终的分析结果,特别是复杂背景下的植物叶片图像分割更是技术的难点与关键。本项目提出基于新型多竞争方式的水平集方法用于多种复杂背景下植物叶片图像分割。主要提出五种新模型：1）使用力矩竞争融合水平集方法来构建能量模型；2）建立高维多尺度自适应的水平集能量模型；3）构建基于局部统计的多特征水平集能量模型；4）构建分层组合的混合竞争方式的水平集能量模型；5) 构建基于稀疏表达融合的水平集能量模型。最后，根据以上五种模型分别对非目标复杂叶片干扰、杂草或光照干扰，复杂光照以及纹理环境干扰，叶片重叠干扰以及同类非目标叶片干扰等复杂背景下的植物叶片图像进行图像分割研究，以便针对不同的叶片图像特点优选出更加合适和鲁棒的分割模型。本项目的深入研究，将对我国正大力倡导的数字化园林植物研究和数字农业起到一定的推动作用。

中文关键词： 植物叶片图像；力矩竞争；高维多尺度；多特征；混合竞争

英文摘要： In the research of digitalized plant leaves, the information extracted and analysis outcomes of plant leaves images shall be influenced by the segmentation methods. Particularly, it is a difficult and key task to segment plant leaves images with complicated background. This project proposes novel multi-mode competition methods based on level set to investigate the segmentation methods of plant leaves images with complicated background. In this project, the five novel research methods are proposed, which include: 1) Energy model method incorporating the moment competition into level set; 2) Energy model method based on the adaptively high-dimensional multi-scale level set; 3) Energy model method based on multi features with local statistical level set; 4) Energy model method incorporating hybrid competition and layered as well as combined level set. 5) Energy model method incorporating sparse representation into level set. Finally, the five novel models will be explored to segment plant leaves images with complicated background so as to testify the effectiveness and robustness of the models. By means of the exhaustive research for this project, the study of digitalized plant and agriculture advocated by our country will be further promoted and developed.

英文关键词： Plant Leaves Image;Moment Competition;High-dimensional Multi-scale;Multi Features;Hybrid Competition