SIA OpenIR  > 光电信息技术研究室
分焦平面偏振成像关键技术研究
Alternative TitleResearch on key techniques of polarization imaging for division-of-focal-plane polarimeters
张俊超1,2
Department光电信息技术研究室
Thesis Advisor罗海波
Keyword分焦平面 偏振成像 图像插值 图像去噪 非均匀性校正
Pages120页
Degree Discipline模式识别与智能系统
Degree Name博士
2019-05-14
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract分焦平面偏振成像是一种新型的偏振信息获取技术,它是将微偏振阵列集成到相机的焦平面上,焦平面上的每个像元对应一个微偏振元,每2*2单元组成一个超像素,按照0°,45°,90°,135°排列。它可以一次成像获得多个偏振方向的光强响应,该系统结构紧凑、体积小,又具有高透过率、高消光比和高实时性等优点,是当前偏振成像的研究热点。然而该成像系统还存在一些成像问题,导致偏振图像质量较低。为此,在本文开展了分焦平面偏振成像关键技术方面的研究,本文的主要贡献如下:1. 针对分焦平面偏振图像分辨率不足和瞬时视场误差问题,提出了一种基于偏振光强相关性的分焦平面偏振图像插值算法。该算法是基于偏振通道之间的相关性和图像的梯度信息进行建模,针对简单的自然场景获得了较好的插值效果,且成功应用于实际工程中。2. 提出了一种基于稀疏编码理论的分焦平面偏振图像去马赛克算法,该方法结合稀疏先验和非局部自相似先验进行建模。模型简单有效,将分焦平面偏振图像插值问题转变成最优化求解问题,通过迭代求解获得高质量的插值图像。实验结果表明该算法相对于传统的插值算法极大程度地提高了插值精度,为复杂场景的偏振图像插值提供了一种有效的技术手段。3. 提出了偏振图像去马赛克的卷积神经网络PDCNN,首次将深度学习应用于偏振成像领域中。PDCNN网络是一个端到端的网络,以马赛克的分焦平面偏振图像作为输入,四个偏振方向的全分辨率图像作为输出。设计了跳跃连接、专门的Stokes矢量层和损失函数,极大程度地提高了偏振图像的插值精度。4. 针对分焦平面偏振图像的非均匀性问题,提出了一种基于标定的可见光波段的非均匀性校正算法。综合考虑了微偏振片和相机探测器的非均匀性,从非均匀性产生机理入手,根据光强获取过程进行建模。并采用了平均分析矩阵求解校正模型,克服了现有校正算法欠校正和过校正的问题。5. 针对分焦平面偏振图像去噪问题,提出了一种基于主成分分析的分焦平面偏振图像去噪算法。该方法直接在分焦平面偏振图像(去马赛克前)上进行去噪,有效地弥补了现有去噪算法的不足。此外,还在Stokes矢量上进行了残差估计和去除,进一步提升了去噪效果。
Other AbstractDivision of focal plane polarimeter is a new method to obtain polarization information. It is composed of a collection of linear polarizers aligned and overlaid upon a focal plane array sensor. Each pixel is aligned to a micro-polarizer. Within 2-by-2 super pixel, four polarization orientations 0°,45°,90°,135° are arranged. This imaging sensor can record intensity measurements at four polarization orientations simultaneously for only one shot. Besides, the system is small in size (compact structure), high in transmittance, high in extinction ratio and high in real-time. It has attracted more attention recently. However, there are still some imaging problems in this imaging system. The quality of reconstructed polarization images is low. To address this problem, the key technologies of polarization imaging for division of focal plane polarimeters are studied in this paper. Our main contributions are as follows: 1. To address insufficient resolution and instantaneous field of view (IFoV) error issue, we propose a correlation-based interpolation algorithm for division of focal plane polarimeters. The interpolation model is built based on the correlation among different polarization channels and gradient information. For simple natural scene, the proposed method outperforms previous methods and it is successfully applied in practical application. 2. A demosaicing algorithm based on sparse representation is proposed. The sparsity and non-local self-similarity priors are integrated into the demosaicing model. This model is simple and effective. It transforms image interpolation question into a optimal problem. By solving minimization problem iteratively, interpolated images are obtained with high quality. Experimental results showed that the proposed method greatly improved the interpolation accuracy and was superior to some state-of-the-art algorithms in terms of quantitative measurement and visual quality. It provides a new way to interpolate full-resolution images under complex scenes. 3. A polarization demosaicing convolution neural network (PDCNN) is proposed. Deep learning is firstly applied in the field of polarization imaging. The PDCNN is an end-to-end network that learns the mapping function between mosaic images and full-resolution ones with four polarization channels. Skip connections and customized Stocks layers and loss function are used to boost the performance. 4. To mitigate non-uniformity errors, a calibration method is proposed in visible waveband. Both the non-uniformity of micro-polarizer and camera detector are considered to build correction model. Starting from the generation mechanism of non-uniformity, the calibration method is derived according to the acquisition process of intensity. The average analysis matrix is used to solve the correction model, which overcomes the under-correction and over-correction problems produced by current calibration approaches. 5. For division of focal plane polarimeters, we propose principle component analysis based denoising method to remove random noise. The denoising is directly carried out on mosaic images, which makes up for the shortcomings of current denoising methods. Besides, residual noise is estimated on the Stokes parameters. The residual noise removal is benefit to denoising results.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/25148
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院大学
Recommended Citation
GB/T 7714
张俊超. 分焦平面偏振成像关键技术研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2019.
Files in This Item:
File Name/Size DocType Version Access License
分焦平面偏振成像关键技术研究.pdf(7552KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[张俊超]'s Articles
Baidu academic
Similar articles in Baidu academic
[张俊超]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[张俊超]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.