高动态成像方法研究
文献类型：学位论文
| 作者 | 侯幸林 |
| 学位类别 | 博士 |
| 答辩日期 | 2017-11-28 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 沈阳 |
| 导师 | 罗海波 |
| 关键词 | 高动态成像，高动态重建，色调映射，图像融合，模糊控制 |
| 其他题名 | High dynamic range imaging method |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 随着计算机技术的快速发展，机器视觉、人工智能已经活跃在人类工作生活中的方方面面，具体有智能交通、智能监控、辅助驾驶等。成像是光电传感器对场景信息的获取过程，是人工智能的“眼睛”，为机器学习等提供不可或缺的数据基础。如何获取能表达场景信息的高质量图像，是学术界和工业界研究的重点和热点问题。高动态场景是亮度范围较大的场景，具体表现为场景中亮度的最大值与最小值的比值很大，而目前的图像获取设备动态范围有限，用低动态范围的相机拍摄高动态场景的信息，往往会使获取的图像质量较低，影响成像结果。高动态成像技术是解决这一问题最有效的方法，目前，针对高动态场景成像的算法主要有以下三方面的难点：1.如何高效获取能表达高动态场景信息的低动态图像序列；2.如何在高动态重建与色调映射的结果中不引入噪声；3.如何在图像融合结果中保留更多的细节。本文将针对高动态成像算法中存在的问题进行研究，具体包括以下几个方面：(1)提出了基于信息熵最大准则的多曝光控制方法和基于模糊逻辑的多曝光控制方法。基于信息熵最大准则的控制方法针对场景中不同的亮度区域，以信息熵最大为准则进行曝光控制，所获取的图像序列可完整表达高动态场景，且获取的图像序列冗余较少。基于模糊逻辑的控制方法以图像的灰度统计信息为优化目标，在设计初始曝光时间的前提下，通过模糊控制的方法快速获取图像序列，相比于基于信息熵最大准则的多曝光控制方法，该方法能在保证图像质量较好的前提下快速获取图像序列，可满足高动态成像的要求。(2)提出了基于多项式拟合的相机响应曲线恢复方法和基于累计分布函数的局部色调映射方法。将相机响应曲线建模为高阶多项式，在此基础上添加约束项，可快速求解多项式系数，恢复相机响应曲线。以灰度的累计分布函数为基础，结合多尺度分解，提出了一种快速局部色调映射方法，该方法可针对高动态图像中的不同细节进行不同的压缩，在不引入噪声的前提下提高了压缩效率。 (3)提出了基于场景分割的高动态成像方法和基于引导滤波的多曝光融合方法。对场景区域进行聚类分割后，在分割的结果中进行最优选取、拼接和平滑操作，可以充分获取场景的有效信息合成结果图。基于引导滤波的多曝光融合方法针对图像序列设计权值，利用加权求和的方法实现图像融合，该方法在不引入噪声的前提下，最大程度地保留图像细节，取得了很好的效果。 |
| 英文摘要 | With the rapid development of computer technology, machine vision and artificial intelligence (AI) have been active in our lives in all aspects of work, including intelligent transportation, auxiliary driving, intelligent monitoring and so on. As the eye of AI, imaging captures the scene information through photoelectric sensors, and provides indispensable data foundation for machine learning. How to obtain high quality images has long been the focus and a hotspot for academia and industry research. For a high dynamic range scene, the radio of the maximum and the minimum brightness is very large. However, the current image acquisition devices have limited dynamic range. The image quality is always low when it is captured by a low dynamic range camera from a high dynamic scene. High dynamic range imaging technology is the most effective way to solve this problem. At present, the following three aspects are the main difficulties for high dynamic range imaging algorithm: (1) How to obtain low dynamic image sequence which could express high dynamic scene effectively; (2) How to reconstruct high dynamic range image and map it to low dynamic range without introducing noises; (3) How to retain more details in the image fusion results. Targeting the problems mentioned above, detailed researches are conducted as follows. (1) Two multi-exposure control methods are proposed based on the principle of maximum informational entropy and the fussy logic algorithm, respectively. The first method controls the exposure time according to the information entropy for different brightness regions. The image sequence could fully express the high dynamic range scene with little redundant images. The second multi-exposure control method considers the gray-scale statistics information of the image as the optimization object. When the initial exposure time is designed, this method could quickly obtain a image sequence through the fuzzy control method. It is highly efficient and could obtain high quality image sequences to meet the requirement of high dynamic imaging. (2) Camera response curve restoration method based on polynomial fitting and local tone mapping method based on cumulative distribution function are proposed. In this paper, the camera response curve is modeled as a higher-order polynomial. On this basis, the constraint items are added to quickly solve the polynomial coefficients, and then the camera response curve is restored. Based on the gray-level cumulative distribution function and multi-scale decomposition, a fast local tone mapping method is proposed. Without increasing noises, this mapping method could differently compress for different regions in high dynamic images with a high compression efficiency. (3) A high dynamic imaging method based on scene segmentation and a multi-exposure fusion method based on guided filtering are proposed. Firstly, the image is segmented into different brightness areas by clustering the neighboring pixels. Secondly, the optimal selection, stitching and smoothing operation are conducted in the segmented images. At last, a result image is fused with using information of the scene. The multi-exposure fusion method based on the guidance filter designs the weights for the image sequence and fuse the images using the method of weighted summation. This method preserves the image details to the maximum degree without introducing noises and achieves good results. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 页码 | 105页 |
| 源URL | [http://ir.sia.cn/handle/173321/21272]  |
| 专题 | 沈阳自动化研究所_光电信息技术研究室 |
| 推荐引用方式 GB/T 7714 | 侯幸林. 高动态成像方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所. 2017. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。