基于散射变换的图像不变特征提取

项目名称： 基于散射变换的图像不变特征提取

项目编号： No.61273244

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： 唐远炎

作者单位： 重庆大学

项目金额： 83万元

中文摘要： 现实环境中距离、视角、光照等变化导致图像产生仿射变形和弹性形变，准确、鲁棒的图像特征提取面临巨大挑战，研究图像不变特征的构造理论和提取算法，对图像的分类和识别具有重大的理论和应用价值。本课题综合考虑仿射形变、弹性形变、光照变化和局部遮挡的影响，采取理论研究与实证研究相结合的方法，紧紧围绕图像不变特征提取问题，开展散射变换理论改进与拓展、散射变换算法快速实现、散射系数的全局统计特征和局部稳定特征提取等三个方面的研究，为图像分析与理解提供理论支撑和方法指导。本课题将在稀疏性复小波的构造、不变性组合散射变换的设计、变换域特性与散射理论的有机结合、信息度量下的散射路径优化、基于散射域空间-尺度模型的图像全局与局部不变性特征提取等方面形成创新和特色，并通过在目标识别和图像分类等方面的应用，验证本课题取得的图像不变特征研究成果。

中文关键词： 散射变换；几何变换；稀疏表示；特征提取；高光谱图像

英文摘要： In real environment, the changes of distance, perspective and light could lead to affine and elastic image deformation, which is the greatest challenge faced for accurate and robust image feature extraction. Discussing the tectonic theory and extraction algorithms of image invariant feature has significantly theoretical and applied value for image classification and recognition. we synthesize the influences of affine deformation, elastic deformation, illumination changes and partial occlusion, combining theoretical and empirical research approach, concentrating on extraction of invariant image feature, this project focuses on improvement and expansion of scattering transform theory, rapid realization of scattering transform algorithm and extraction of global statistical characteristics and local stable feature of scattering coefficient, which would provide theoretical support and methodological guidance for image analysis and understanding. The innovations of this project would include: construction of sparse complex wavelet, designation of invariant combining scattering transform, organic integration of domain characteristics and scattering theory, information measure based scattering path optimization, global and invariant local image feature extraction based on scattering space-scale model. The achievements o

英文关键词： Scattering transformation；geometry transformation；sparse representation；feature extraction ；hyperspectral image