SGER: Computational Photography: Lighting and Focus

智力优势：拟议的研究计划将结合计算机图形学和摄影的概念，借鉴PI在这两个领域的广泛经验。其基本概念是将计算机图形技术应用于摄影，以解决摄影过程中的根本缺陷。这涉及到将算法开发的思路扩展到物理可实现的范围之外。方法：对于光照不均匀的集体照片，我们将开发亮度补偿技术来修改曝光的肖像，使每个人脸都具有大致相同的亮度。该方法提取曝光不足的人脸区域及其平均亮度值和位置，然后根据被摄体布局的分类添加亮度来补偿它们的曝光不足。该方法将涉及一系列图像处理步骤：提取肤色对象和面部区域，提取并确定布局分类，利用平均亮度和空间变化添加亮度。为了实现摄影中对焦的普遍控制，我们提出了一种合并和处理照片的算法。一种方法是首先拍摄多张照片，每张照片都有不同的对焦设置，然后将这些图像组合在一起，生成只有所需区域对焦的合成图像。在我们的软件控制下，使用数码相机可以用精确校准的对焦设置快速拍摄许多不同的照片。对于运动场景，如果多次曝光，在不同的曝光下会出现不同的效果，我们提出了一种不同的方法，综合添加模糊作为后处理，以接近期望的效果。影响：本研究的结果可能在数码摄影的各个方面都有应用，包括相机和软件，并将影响到新兴的和目前处于起步阶段的计算摄影领域。我们的算法最初将作为后处理在计算机上作为软件原型，但最终我们希望看到它们集成到数码相机的固件中。

Intellectual merit: The proposed research program will integrate concepts from both computer graphics and photography, drawing on the extensive experience of the PI in both endeavors. The underlying notion is the application of computer graphics techniques to photography to solve fundamental shortcomings in the photographic process. This involves extending our thinking for algorithm development beyond what is physically realizable.Methods: For a group photograph that has non-uniform illumination, we will develop brightnesscompensation techniques to modify an exposed portrait such that every face would have approximately the same brightness. Our approach will extract underexposed facial regions with their average brightness values and locations, and then compensate for their underexposure by adding brightness according to the classification of the layout of the subjects. The proposed method will involve a sequence of image processing steps: extracting skin-colored objects and facial regions, extracting and determining a layout classification, and adding brightness using average brightness and spatial variation.To achieve generalized control over focus in photography, we propose to develop algorithms that will combine and process photographs. One approach will involve first taking many pictures, each with different focus settings, and then combining these images to produce a composite in which only the desired region is in focus. It will be possible to quickly take many different pictures with precisely calibrated focus settings using a digital camera under the control of our software. In the case of a moving scene that would appear different in different exposures if multiple exposures were taken, we propose a different approach that synthetically adds blur as a postprocess, to closely approximate the desired effect.Impact: The results of this research could have applications in various aspects of digital photography, both for cameras and software, and will impact the nascent and currently inchoate field of computational photography. Our algorithms will initially be prototyped as software on a computer as a postprocess, but eventually we hope to see them integrated into the firmware of digital cameras.