项目名称: 基于图像的植物识别和检索研究
项目编号: No.61272285
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 范建平
作者单位: 西北大学
项目金额: 82万元
中文摘要: 随着信息技术的快速发展,人们可以很便捷地获取喜爱植物的照片,并期望能在互联网上查找到该未知植物的相关信息。虽然基于关键词的文本搜索技术已相对成熟,但人们在多数情况下并不知道所感兴趣植物的名称,因此找不到合适的查询词。为满足人们日益增长的鉴赏和了解各类植物的需求,本课题将研究一种新的基于图像匹配和统计学习算法的大规模植物图像的自动识别和检索框架:1)基于大量植物类网站来全自动地构建大规模植物图像库;2)结合相机的元数据和植物图像的视觉特征来更准确地检测图像感兴趣区域;3)基于植物分类学和植物特征建立大规模植物概念网络;4)基于植物概念网络的多任务学习和多层次学习算法,利用植物概念之间的语义和视觉关联性来指导训练大规模植物概念分类器;5)基于自适应的图像摘要和交互式可视化的植物概念分类器评估方法。针对植物识别与检索这一特殊研究主题的深入探索,其成果也将推动图像处理与检索等相关领域的研究和发展。
中文关键词: 植物识别;基于内容的图像检索;植物概念网络;分层学习;多任务学习
英文摘要: With the development of digital camera, smart phone and Internet search technology, we can easily capture the photos of the plants of interest and expect to search the useful information about the unknown plants. Even there has significant progress on keyword-based text retrieval, it is still very hard for people to find suitable keywords to enable keyword-based retrieval of the relevant information for unknown plants. With new requirements from users on unknown plants, this project will integrate image matching and statistical learning algorithms to enable automatic image-based plant recognition and retrieval: 1) leveraging large number of plant Web sites to crawl large-scale plant images; 2) integrating camera meta data with the visual properties of plant images to detect the regions of interest automatically; 3) leveraging domain knowledge and visual properties of plant images to construct a plant concept network; 4) incorporating the plant concept network, multi-task learning and multi-level boosting to train a large number of plant concept classifiers with higher discrimination power; 5) integrating automatic image summarization and interactive visualization to assess the effectiveness and correctness of plant concept classifiers. This project will focus on automatic plant concept recognition and retrieval,
英文关键词: Plant Identification;Content-based Image Retrieval;Plant Concept Network;Hierarchical Learning;Multi-task Learning