图像语义分析是物体识别与图像内容理解中的关键技术。本项目将在模式识别、机器学习、计算机视觉等交叉领域最新研究进展基础上，针对稀疏表示和正则化技术应用到图像语义分析所面临的一些关键问题，深入研究其解决方案。基于最小描述长度（MDL）原理框架，研究稀疏建模中的正则化技术和正则化参数选择方法；在此基础上，结合标记样本的判别信息和字典的内部性质，构建适用于图像语义分析的稀疏模型；以及发展基于混合矩阵正态分布的图像语义分析模型。通过集成相关研究成果和技术，完善图像语义分析原型系统，验证所发展的理论和算法的有效性。

Image semantic analysis is the key technologies for object recognition and image understanding. We will base on the latest research advances in the cross fileds of pattern recognition, machine learning, computer vision, focus on some key problems of sparse representation and regularization method application to image semantic analysis, investigates the way to find the solutions in this project. The regularization techinique and the method of regularization parameter selection in the sparse modeling will be studied under the framework of Minimal descrition length (MDL) principle; On the basis of these studies, the sparse models are constructed by combining the discriminitive information and inner properties of dictionary and image semantic analysis model based on mixture matrix normal distribution is developed.Finally, a prototype system of image semantic analysis will be developed further by integrating the related reaearch results and technique, and the system will be used to verify the effectiveness of developed theory as well as algorithms.