视觉质量感知的脑电时空特性研究

This project, aiming at solving the problem that the otherness among individual cognitions and the instability of the results in subjective quality assessment, explores the quality assessment method to obtain the visual perceptual quality through analysis the spatio-temporal correlation of electroencephalography (EEG) signal which contains abundant physiological and psychological information. This project is implemented by designing the pattern of perceptual assessment to capture effective EEG signal, constructing learning model to extract feature via keeping the EEG’s spatio-temporal correlation structure, and building criterion to quantitatively describe the relationship between the EEG and perceptual quality. The main research contents include: (1) design the test pattern of the visual quality assessment to capture EEG signal through constructing the mode of stimulation, designing the pattern of perceptual test, and modeling the form of cognitive response. (2) analysis the spatio-temporal characteristic of EEG signal through studying the spatio-temporal correlation of multi-channel EEGs, and extracting the characteristics which can represent visual perception quality effectively; (3) build the relationship model between the features of EEG signal and perceptual quality through studying the law of EEG changes as perceptual quality, designing the criterion of quantitative description, and modeling the mapping relationship between the EEG and perceptual quality. In summary, based on the physiological mechanisms of EEG, a new framework of visual perceptual assessment is constructed via combining with low-rank representation and deep learning. It can provide some new ideas for image quality assessment theory, and also provide the theoretical basis and technical support for studying the perception and cognitive mechanism of the human visual system.

本项目旨在针对视觉质量主观评价中的个体认知差异性和结果不稳定性问题，基于脑电信号中蕴含的生理和心理信息，分析脑电信号的时空特性，探索视觉感知质量的获取方式。本项目拟通过设计有效捕获脑电信号的感知评价范式，构建保持脑电时空相关结构的特征学习模型，建立脑电信号与感知质量关系的定量描述准则，主要研究：（1）视觉质量评价感知脑电获取的实验范式设计，研究事件呈现模式的构建、感知测试范式的设计和认知响应形式的模拟；（2）感知质量脑电信号的时空特性分析，研究多通道脑电信号时空相关关系，提取有效表征视觉感知质量的特征；（3）脑电信号特征与感知质量关系模型的构建，研究脑电随着感知质量变化的规律，定量描述准则的设计和映射关系的建模。本项目基于感知脑电的生理学特性，结合低秩表示和深度学习等最新机器学习方法，构建视觉感知质量评价的新框架，丰富和完善质量评价理论，同时为视觉信息的感知认知机理提供理论依据和技术支撑。

视觉质量感知是人类认知机理的基本问题之一，具有重要的理论意义和应用价值。本项目通过直接测定感知过程中的脑电信号，分析其时空特性，构建脑电信号与感知质量之间的关系，探索质量感知过程中的生理和心理机制，并将感知特性应用于客观质量评价方法的设计和视觉信息的增强。本项目在以下几个方面取得重要进展和成果：（1）研究了基于脑电信号的视觉特性和感知质量之间的关系。设计了基于Oddball范式的主观质量评价实验，对采集后的脑电信号进行时空特性分析，发现对于不同质量图像的ERP波形的响应程度不一样，不同失真类型图像激活的脑区具有显著差异性。设计了基于Go-NoGo范式的多组不同美观度的评测实验，诱发不同的事件相关电位差异波形，验证了人脑对图像美感度的认知过程可定量测量。同时，利用深度神经网络模型描述脑电信号特征与感知质量之间的关系，使其能有效、定量地描述质量感知的机制。（2）研究了基于感知特性的视觉信息质量评价方法。基于人类对于色彩感知的不均匀性和视频运动的整体性，提出了一系列视觉信息质量的评价方法，取得较好的主客观一致性。（3）研究了基于视觉感知特性的质量提升与认知技术。利用视觉感知的稀疏性、视频序列的连续性和视觉认知的一致性等特性，提出了图像质量增强方法和视觉信息识别技术，有效地扩充了视觉信息的感知-认知机理。