基于自然图像复杂视觉信息的特征提取算法与应用  

FEATURE EXTRACTION ALGORITHM BASED ON COMPLEX VISUAL INFORMATION OF NATURAL IMAGE AND ITS APPLICATION

作　　者：赵彦明[1] 房健[1] 季圣杰 

机构地区：[1]河北民族师范学院,河北承德067000 [2]昆明理工大学,云南昆明650093

出　　处：《计算机应用与软件》2015年第11期200-205,共6页Computer Applications and Software

基　　金：河北省教育厅科学技术研究项目(Z2014187)

摘　　要：分析自然图像复杂视觉拓扑组织结构的邻域关系生成树表示方法的缺欠,提出基于自然图像复杂视觉信息的特征提取算法与应用。算法定义通用视觉细胞感受野模型,提出自然图像复杂视觉结构信息的超完备基邻域关系生成树表示方法;并以自然图像的慢变特征和中心极限定理为理论基础,改进准正交超完备基预测算法,自适应地构造自然图像超完备基邻域关系生成树;遍历超完备基生成树,计算自然图像与节点的最大响应系数,实现树形拓扑组织结构编码,并采用海明距离计算编码相似度,实现自然图像分类。实验表明:自然图像集中,该算法学习的基滤波器集具有类似于V1区复杂视觉细胞的视觉特征;对输入图像的局部变化具有良好的几何不变性;在北卡莱罗那州立大学提供的自然图像图库上,进行基于内容的图像检索比较实验,该算法比传统视觉模型具有更好的检索效果和抗噪声能力。We analyse the shortcomings of the representation of neighbourhood relation spanning tree of complex visual topology organisation structure of natural images, and propose the natural image complex visual information-based feature extraction algorithm and its application. The algorithm proposed in the paper defines the general receptive field model of visual ceils, and presents the representation approach for overcomplete base neighbourhood relation spanning tree of complex visual structure information of natural images. Moreover it improves the prediction algorithm of quasi-orthogonal complete base by taking the slowly vmTing feature of natural image and central limit theorem as the theoretical bases, adaptively constructs the overcomplete base neighbourhood relation spanning tree of natural images; by traversing the overcomplete base spanning tree, it calculates the maximum response factor of natural image and nodes, realises tree topology organisation structure coding. Furthermore, it uses hamming distance to calculate the similarity of coding to implement the classification of natural images. Experiments show that in natural images set, the base filter set learned by the algorithm in the paper has visual features similar to that of the complex visual cells in V1 area; it has good geometrical invariance on local changes of the inputted image. On natural image gallery provided by the North Carolina State University we carried out the content-based image retrieval comparative experiment, the proposed algorithm has better retrieval performance and noise immunity than the traditional visual model.

关 键 词：拓扑结构 超完备基 感受野 海明距离 几何不变性 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]