基于视觉感知的场景再现技术研究

With the development of computer and multimedia technology, medical imaging, video surveillance, satellite remote sensing and other various fields such as computer vision require higher quality image. High quality images can provide more abundant information and more real visual perception to human eyes. In order to obtain the high quality images, we will do some research on scene rendition based on vision perception theory in the project. It includes four main contents:.(1)Research on the tone mapping based on human visual perception. As one of the most basic and important visual functions, visual perception for human vision system plays a key role for the human beings stably perceiving the real world. In order to make computer vision system have the similar visual perception, in this project, perception theory is introduced in the tone mapping methods for displaying HDR images..(2)In most exposure fusion algorithms in the spatial domain, single image feature is usually served as a standard to evaluate the quality of the image. However, the single feature is not comprehensive. So, the relationship between the multi- dimension image characteristics and the image quality is a key point for fusion rule design in the exposure fusion methods.In this project, we propose a new method to construct the relationship between image characteristics and the image quality..(3)For color constancy under uniform illumination, which is supervised by high-level visual information, there is little work on this direction. Now only indoor-outdoor information is utilized. To overcome such shortcoming, in the project, a much more generalized scene classification method is introduced, and the high-level semantic information is used in the classification method, which can increase the accuracy of the illumination estimation. .(4)When abrupt illumination change exists in the scene, illuminant estimation is difficult compared to single illumination color constancy methods. Most methods are generally based on the simplifying assumption that the spectral distribution of the light source is uniform across the scene. However, in reality, this assumption is often violated due to the presence of multiple light sources. So how to estimate light source is key issue for the color constancy.?In this project, we propose a new method to estimate the light source, and a color constancy for multiple illuminants is proposed.

高质量并符合人类视觉感受的图像能够提供更丰富的信息，是很多实际应用的基础。本项目以人类视觉系统感知为研究背景，通过对HDR图像处理和颜色恒常性计算进行研究，使获取的图像更加符合人眼对真实场景的感知。主要内容包括：（1）基于HVS视觉感知的高动态范围图像色调映射算法研究；（2）针对多输入图像，研究多维图像特征和图像质量衡量标准的映射关系，并建立具有高鲁棒性的融合规则；（3）在现有的底层特征驱动的颜色恒常性计算基础上，研究融合高层语义信息来指导单光照颜色恒常性计算；（4）从突变光照角度出发，研究多光照下颜色恒常性计算。真实场景再现技术能够使自然场景在普通显示器得到逼真地再现，这种效果在高清影片、游戏特效、印刷等领域都有广泛的应用，所以它不仅具有非常重要的学术价值而且具有较高经济和社会价值。

由于人类对图像的视觉质量要求不断提高，特别是医疗影像、高清影片、游戏特效、航空航天等领域的需求更为突出，本项目以视觉感知理论为基础,主要研究基于颜色感知计算的场景可视化技术，包括多曝光融合和颜色恒常性计算两方面，研究成果能够使处理后的图像在显示设备上的显示效果符合人类视觉系统对真实场景的客观感知。主要研究内容包括：（1）提出一种利用高层场景相似度指导下的有监督的颜色恒常性算法。（2）通过分析过完备稀疏表示的特点，提出一种基于稀疏表示多曝光融合算法，并设计一种基于稀疏度的融合规则，可保留场景更多的细节信息。（3）针对低质图像，提出一种基于低秩矩阵稀疏分解的图像复原方法。本项目的研究成果对人脸识别、肤色检测、视频监控等领域都具有深远影响，该课题不仅具有非常重要的学术价值而且具有较高经济和社会价值。.经过三年的研究，本项目主要获得了以下科研成果：出版学术专著1项，申请专利2项，软件著作权2项；发表学术论文16篇，其中SCI收录3篇，国际顶级会议CVPR 1篇，ECCV 1篇；培养硕士研究生2名；晋升高级职称的教师1名；参加国际学术会议2次。