稀疏二值图像特征提取的数据场方法  

Data Field-based Feature Extraction Method for Sparse Binary Image

作　　者：吴涛[1] 陈一祥[2] 杨俊杰[1] 

机构地区：[1]湛江师范学院信息学院,湛江524048 [2]武汉大学遥感信息工程学院,武汉430079

出　　处：《计算机科学》2014年第10期310-316,共7页Computer Science

基　　金：广东省自然科学基金(S2013040014926;S2012010009759);广东高校优秀青年创新人才培养计划(2012LYM_0092);湛江师范学院科学研究项目博士专项(ZL1301);国家973重点基础研究发展计划(2012CB719903)等资助

摘　　要：针对图像特征自动提取问题,以稀疏二值图像为例提出了一种模拟物理学场论机制的数据场方法。该方法首先建立二值图像数据场实现图像特征空间到数据场势值空间的映射关系,搜索每个非零像素的八连通区域;然后计算势值及其主方向角,生成势值矩阵和方向角矩阵;最后通过势值归一化和主方向归一化输出以势值和主方向为基础的特征向量及其对应的可视化曲线。新方法利用数据场解决图像特征提取问题,能兼顾图像灰度空间的局部性和数据场势值空间的全局性。手写数字图像的定性和定量实验表明,该方法特征提取效果较好、性能稳定,具有合理性和有效性。In order to extract the image feature automatically, a novel data field-based method for sparse binary image was proposed from the point of view of the physics-like field theory. First, the method constructs a map from grayscale space to potential space by producing the data field for a given binary image. Next, it calculates the potential value and the principal direction for each pixel with non-zero value by scanning its 8-connected regions, and then obtains the potential matrix and the direction angle matrix. Finally, it generates the feature vectors and its corresponding visual curve after the normalization of potential value and principal direction. The proposed method solves the issue on image feature extraction using data field,and it can keep a balance between the locality of image grayscale space and the globality of potential space in data field. The quantitative and qualitative experiments with the handwritten digital images indicate that the proposed method yields accurate and robust feature extraction results, and is reasonable and effective.

关 键 词：数据场 认知物理学 图像特征 二值图像 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]