场景内、外物体与场景的关联机制研究

发布时间：2018-06-10 03:58

本文选题：LOC + PPA　；参考：《天津大学》2016年硕士论文

【摘要】：利用计算机视觉技术看懂图像仍是一件非常复杂的事情,而利用计算机完成对场景的识别更是一项极具困难的挑战,但人脑却能够轻松看懂图片、识别场景。因此,通过借鉴人脑的视觉信息加工机制,为计算机构建模拟人类视觉和脑部工作机制的算法和模型,将有助于推动计算机视觉领域的发展和进步。本文旨在通过研究人脑视觉场景认知过程中,场景内、外物体和场景之间的关联机制,为计算机视觉的发展提供理论和方法上的依据。本文使用四类场景图片(包括两类室内场景和两类室外场景)、每类场景内的两个显著性物体图片、不在场景中出现但和场景语义相关的两个物体图片作为刺激材料。利用功能磁共振(fMRI)记录受试在观看这些刺激材料时的大脑活动信号。通过对场景内、外物体和场景的激活模式在不同感兴趣区域进行多体素模式相关分析和回归分析发现,在LOC(两侧枕部复杂区域)的两个子区LO(两侧枕部区域)、pF(梭状回后部区域)和PPA(海马旁回区域)脑区,物体与场景的激活模式显著正相关,且场景外物体与场景的关联强度显著小于场景内物体与场景的关联强度。对比脑区之间关联强度发现LO和p F脑区所表征的关联强度显著强于PPA脑区。进一步分析发现,LO和pF表征的物体与场景关联强度对于室内、室外场景存在差异,PPA区域没发现该差异。上述发现在RSC(扣带回皮层)脑区域均没有观察到。基于上述结果推测,LO、pF和PPA脑区参与了场景内、外物体与场景相关联的信息加工,共同完成物体与场景关联关系处理的过程。其中LO和pF主要的表征语义关联,PPA主要表征场景空间特征和语义关联,且均对场景的开阔程度敏感,RSC不表征场景与物体的关联关系。
[Abstract]:Using computer vision technology to understand images is still a very complicated task, and it is a very difficult challenge to use computer to realize the recognition of scenes, but the human brain can easily understand pictures and recognize scenes. Therefore, it is helpful to promote the development and progress of computer vision field by drawing lessons from human brain visual information processing mechanism and constructing algorithms and models for computer to simulate human vision and brain working mechanism. The purpose of this paper is to provide theoretical and methodological basis for the development of computer vision by studying the correlation mechanism between external objects and scenes in the cognitive process of human visual scene. In this paper, four kinds of scene pictures (including two kinds of indoor scenes and two kinds of outdoor scenes) are used as stimulus materials. Two images of significant objects in each type of scene are used as stimulative materials, which are not present in the scene but are related to the scene semantics. Functional Magnetic Resonance Imaging (fMRI) was used to record brain activity during viewing these stimuli. Based on the correlation analysis and regression analysis of multiple voxel patterns in different regions of interest, the activation patterns of external objects and scenes in the scene were analyzed and analyzed by regression analysis. In the two subregions of LOC (bilateral occipital complex region) (bilateral occipital region, posterior region of fusiform gyrus) and PPA (region of parahippocampal gyrus), there was a significant positive correlation between object and scene activation mode. And the correlation intensity between objects and scenes is significantly lower than that between objects and scenes. Compared with the correlation intensity between brain regions, it was found that the correlation intensity of Lo and PF brain regions was significantly stronger than that of PPA brain regions. Further analysis found that the correlation strength between objects and scenes represented by Lo and PF was different in indoor and outdoor scenes but not found in PPA area. None of these findings were observed in the RSC (cingulate cortex) brain region. Based on the above results, it is inferred that the brain regions of LOPF and PPA are involved in the processing of the information associated with the scene and the external objects are involved in the processing of the correlation relationship between the objects and the scene, and the process of processing the relationship between the objects and the scene is completed together. Among them, Lo and PF mainly represent semantic association and PPA mainly represent spatial feature and semantic association of scene, and both are sensitive to the open degree of scene. RSC does not represent the relation between scene and object.
【学位授予单位】：天津大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】