一种基于图像固有维度的感知物体检测方法  

Perceptual object detection algorithm based on image intrinsic dimensionality

作　　者：邵静[1] 高隽[1] 赵莹[1] 张旭东[1] 

机构地区：[1]合肥工业大学计算机与信息学院,合肥230009

出　　处：《仪器仪表学报》2008年第4期810-815,共6页Chinese Journal of Scientific Instrument

基　　金：新世纪优秀人才支持计划(NCET-04-0560)资助项目

摘　　要：越来越多的心理学行为实验结果支持了基于物体的视觉注意,感知物体的定义和检测是建立基于物体的视觉注意计算模型的前提和关键。本文提出一种基于图像固有维度的感知物体检测算法。结合初期特征分析理论和拓扑性质感知理论的研究结果,由边缘、同质性区域和显著角点定义感知物体。引入图像固有维度的概念及其一种连续形式定义,利用结构张量法度量图像区域的固有维度,并由此检测出图像中的感知物体。感知物体的检测为基于物体的视觉注意计算模型提供了先决条件,可应用于目标识别、图像分割和场景分析中。对真实图像的仿真实验验证了该算法的合理性和有效性。Object-based visual attention is overwhelmingl results. The definition and detection of perceptual objects y demonstrated by recent psychology behavior experiment is the key issue for building the object-based visual selective attention computational model. A perceptual object detection algorithm is proposed in the paper. Considering the achievement of primary feature analysis theory and topological perceptual theory, the perceptual object is defined through the salient point, edge and homogeneity area. The image intrinsic dimensionality is introduced and a continuous definition of it is presented. The intrinsic dimensionality of the image patch is measured by the structure tensor method, and then the perceptual object is detected. The definition of the perceptual object has provided a precondition for the object-based visual attention computational model. The result can be applied to object recognition, image segmentation and scene analysis. The rationality and validity of the proposed algorithm is validated from great number of experiments on real images.

关 键 词：视觉注意 选择基元 感知物体 图像固有维度 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]