自然场景脑认知图像编解码特征提取与识别技术研究

Though human brain is very efficient and robust when processing the complex visual information, very little has been known about the neural mechanism related with the encoding and decoding procedures of the natural scenes. The development of the magnetic resonance imaging (MRI) technology and corresponding methodologies provide us with a new way of studying brain encoding and decoding. However, many key techniques need further breakthroughs and developments. The present study is thus directed towards the issues arising from the encoding and decoding of brain-cognitive imaging with natural scenes. We attempt to address the critical technical issues involved in the functional MRI-based extraction of representative information from the encoding and decoding of brain-cognitive imaging and the related pattern recognition. These issues mainly include: developing the topological feature extraction methods for brain cognitive mode networks of natural scenes based on MRI, detecting the key brain regions for encoding visual scenes, and finding the corresponding activity and network topological features; establishing the mathematical mapping models between the visual scene stimulus and brain imaging of cognitive activities, and revealing the information features of natural scenes in the brain neural imaging and the spatiotemporal relationships; developing multi-voxel pattern recognition methods for decoding of brain activities, and performing the pattern classification of different natural scene images according to the encoding information embodied in the brain activities; establishing the pattern recognition methods of the visual scenes and dynamical brain cognitive imaging, and revealing the brain comprehension and cognition procedures. This study can provide new means for the feature extraction and recognition of visual images and brain cognitive activities with dynamic movie, and offer new ideas and key technologies for the reading and understanding of human brain.

人脑对复杂视觉信息处理具有高效、鲁棒等优点，但对自然场景认知编解码的神经机制依然知之甚少。随着磁共振成像技术和相应方法学的发展，给研究脑编解码提供了新手段，但仍有许多关键技术需要突破和发展。项目针对自然场景脑认知图像编解码问题，解决利用功能磁共振成像时认知图像编解码脑功能信息特征提取与识别的关键技术问题。主要有：发展基于磁共振成像的自然场景脑认知活动网络拓扑特征信息提取方法，探测编码视觉场景的关键脑区及其对应的活动特征和网络拓扑特征信息；建立视觉场景激励与脑认知影像间的数学映射模型，揭示自然场景在脑神经影像上的功能信息特征与时空关联关系；发展多体素模式识别方法进行脑活动信息解码，实现根据脑活动编码信息对不同自然场景图像的模式分类；建立视觉场景和动态脑认知的模式识别方法，揭示脑理解与认知过程。项目将为视觉图像和动态电影脑认知活动的特征提取与识别提供新的手段、为解读脑带来新思路和关键技术。

人脑对复杂视觉信息处理具有高效、鲁棒等优点，但对自然场景认知编解码的神经机制依然知之甚少。随着磁共振成像技术和相应方法学的发展，给研究脑编解码提供了新手段，但仍有许多关键技术需要突破和发展。项目针对自然场景脑认知图像编解码问题，解决了利用功能磁共振成像时认知图像编解码脑功能信息特征提取与识别的关键技术问题。主要有：发展了基于磁共振成像的自然场景脑认知活动网络拓扑特征信息提取方法，探测了编码视觉场景的关键脑区及其对应的活动特征和网络拓扑特征信息；建立了视觉场景激励与脑认知影像间的数学映射模型，揭示了自然场景在脑神经影像上的功能信息特征与时空关联关系；项目发展了多体素模式识别方法进行脑活动信息解码，实现了根据脑活动编码信息对不同自然场景图像的模式分类；建立了视觉场景脑认知的模式识别方法，揭示了脑理解与认知过程等等这些研究成果在本项目资助下已经在国际上发表SCI科研论文13篇，项目的部分研究成果已获得相关授权国家发明专利2项。项目作出的研究将为视觉图像脑认知活动的特征提取与识别提供新的手段、为解读脑带来新思路和关键技术。同时,本项目的开展培养了多名高校青年教师和博士/硕士研究生。