折反射散焦模糊图像复原理论与方法研究

The defocus blur problem of catadioptric omni-directional imaging, which is caused by mirror curvature, becomes more severely along with introducing high resolution sensors and large aperture. Blur kernel estimation and omni-directional imaging prior can be considered as two main problems in the topic of the image restoration. This project studies the deblurring algorithms of catadioptric omni-directional imaging system, for the purpose of obtaining sharp omni-directional image. The study content contains: (1) Investigate a mathematical restoration model for omni-directional imaging which includes related factors that cause the defocus blur. (2) Propose catadioptric constrains in the process of the blur kernel estimation under the frame of MAP blind deblurring framework, which helps us to iteratively calculate the correct spatial variant point spread function. (3) Introduce an omni-directional image prior which can be used for the optimization of image blur restoration. (4) Present an omni-directional residual guided approach for handling ringing effect. These three aspects can be considered as innovative algorithms that are specially designed for the omni-directional image deblurring by analyzing the catadioptric camera system. This research has an important meaning in improving catadioptric omni-directional imaging quality and promoting applications in related fields.

曲面光学结构造成的散焦模糊是折反射全向成像中不可避免的问题。解决这个问题的主要难点在于：全向图像散焦模糊核的估计与全向图像先验知识的描述。本项目将针对以上技术难点，研究折反射散焦模糊图像复原的理论与方法，目的是得到高清晰的全向图像。研究内容：(1)折反射成像的散焦模糊数学模型；(2)结合全向图像特点的模糊核估计方法；(3)符合全向图像先验的正则化方法；(4)适用于全向图像的振铃抑制算法。特点与创新是：充分考虑和分析折反射成像的模糊成因，根据其特有的曲面反射光学成像原理，对全向图像实施特定优化算法进行图像复原。研究成果对提高折反射全向图像质量，促进其在相关领域的广泛应用具有重要意义。

曲面光学结构造成的散焦模糊是折反射全向成像中不可避免的问题。解决这个问题的关键点在于：全向图像散焦模糊核的估计与全向图像先验知识的描述。本项目针对这个问题，研究折反射散焦模糊图像复原的理论与方法，得到高清晰全聚焦的全向图像。具体的，开展了以下几个方面的研究：(1) 符合全向图像先验的正则化方法和图像复原技术；(2) 图像模糊程度度量和标定技术；(3) 规避散焦模糊的多镜面反射式成像技术；(4) 基于深度学习的端到端图像复原技术；(5) 基于计算摄影的模糊图像增强和优化技术。本课题充分考虑和分析折反射成像的模糊成因，根据其特有的曲面反射光学成像原理，对全向图像实施特定优化算法进行图像复原。研究成果提高了折反射全向图像质量，促进其在相关领域的广泛应用。