基于MDL原理的图像语义特征分析方法研究

项目名称： 基于MDL原理的图像语义特征分析方法研究

项目编号： No.90820010

项目类型： 专项基金项目

立项/批准年度： 2009

项目学科： 轻工业、手工业

项目作者： 郭平

作者单位： 北京师范大学

项目金额： 50万元

中文摘要： 图像语义特征分析是物体识别与图像内容理解中的关键技术。本课题将结合计算智能技术，基于最小描述长度（MDL）原理框架, 深入研究图像语义特征提取所面临的关键性技术问题的解决方案，发展图像语义特征分析的新理论与方法。主要研究内容包括：图像低层视觉特征表述与融合、图像分类模型及优化、图像语义表述模型等。在上述研究基础上，提出并实现图像语义特征分析的高效计算模型，解决图像低层视觉特征选择与融合算法，以及特征表述向量构建、图像贝叶斯分类模型的泛化能力与正则化参数估计、图像高层语义表示的自适应粒度模型设计与构建等关键问题，获取创新性的研究成果。在视听觉信息处理的基础理论研究方面取得进展，为视听觉信息处理在国家安全和社会发展应用提供相关技术支撑。

中文关键词： 最小描述长度；粒度计算；图像语义；计算智能；贝叶斯分类器

英文摘要： Image semantic feature analysis is the key technologies for object recognition and image understanding. This project will employ computational intelligence technology, under the framework of the minimum description length (MDL) principle, study it in depth for finding the solution of some key technical issues encountered in image semantic feature extraction, and develop new theories and methods for .image semantic feature analysis. The main research topic include: low-level visual features of image interpretation and fusion, image classification model and optimization, image semantic interpretation model and so on. Based on the above-mentioned study, we intend to propose and implement high-performance computing model of image semantic feature analysis, to develop the image low-level visual feature selection and fusion algorithm, to solve these key problems such as in construction of feature representation vector, the generalization ability of image Bayesian classification model, the regularization parameter estimation, to design and construct the adaptive granular model for image high-level semantic representation and so on, expect to obtain innovative research results. The goal of ours it to make progress in fundamental theory of visual and auditory information processing, and to provide the relevant technical support in the national security and social development applications.

英文关键词： Minimum Description Length; Granular Computing; Image Semantics; Computational Intelligence; Beyesian Classifier